Recourse and Residential Mortgage Default: Evidence from US States Andra C. Ghent Department of Real Estate, Baruch College, City University of New York Marianna Kudlyak Research Department, Federal Reserve Bank of Richmond US mortgages are commonly assumed to be non-recourse loans, either by law or in practice. Feldstein (2008), for example, asserts that “mortgages are generally ‘no recourse’ loans” and suggests that the reason Europe has much lower default rates is because mortgages in Europe are recourse loans. Those who do recognize that recourse is available in some states still do not attribute a significant role to recourse in the borrower’s default decision. For example, Karl E. Case points out that lenders rarely pursue recourse: “It’s tough to do recourse. It’s costly, and the amount of people’s non-housing wealth tends to be pretty slim” (Leland 2008). A popular real estate textbook states that “[t]he value of deficiency judgments is always open to serious question. This is true in part because of the ways by which they can be avoided or defeated” (Brueggeman We thank the editor (Matt Spiegel), two anonymous referees, Brent Ambrose, Larry Cordell, Jane Dokko, Kris Gerardi, Jeffrey Gunther, Bob Hunt, Ned Prescott, Peter B. Ritz, Brent Smith, Jay Weiser, and Paul Willen, as well as participants at the 2010 ALEA Annual Meeting, the 2010 AREUEA Mid-Year Meeting, seminars at FRB Richmond and FRB Philadelphia, the Federal Reserve System Committee on Financial Structure and Regulation Meeting, the FDIC/FHFA Symposium on Assessing Mortgage Default Risk, the University of Wisconsin– FRB Atlanta HULM Conference, the 2010 SED meetings, and the Wegmans Conference at the University of Rochester for their comments. We thank Mark A. Watson at FRB Kansas City for assistance with the data and Anne Davlin for excellent research assistance. Andra Ghent gratefully acknowledges funding from the PSCCUNY Research Foundation under grant 63044-00-41. The views expressed in this article do not necessarily reflect those of the Federal Reserve Bank of Richmond or the Federal Reserve System. All errors and omissions are the authors alone. An earlier draft of this article circulated under the title “Recourse and Residential Mortgage Default: Theory and Evidence from US States.” This work was supported by the Research Foundation of the Professional Staff Congress of the City University of New York [grant number 63044-00-41]. Send correspondence to Andra C. Ghent, Department of Real Estate, Baruch College, City University of New York, 137 E. 22nd St. Rm. 406, New York, NY 10010; telephone: 646-660-6929. E-mail: andra.ghent@baruch.cuny.edu. c The Author 2011. Published by Oxford University Press on behalf of The Society for Financial Studies. All rights reserved. For Permissions, please e-mail: journals.permissions@oup.com. doi:10.1093/rfs/hhr055 Advance Access publication June 29, 2011 Downloaded from rfs.oxfordjournals.org by guest on September 7, 2011 We quantify the effect of recourse on default and find that recourse affects default by lowering the borrower’s sensitivity to negative equity. At the mean value of the default option for defaulted loans, borrowers are 30% more likely to default in non-recourse states. Furthermore, for homes appraised at $500,000 to $750,000, borrowers are twice as likely to default in non-recourse states. We also find that defaults are more likely to occur through a lender-friendly procedure, such as a deed in lieu, in states that allow deficiency judgments. We find no evidence that mortgage interest rates are lower in recourse states. (JEL E44, G21, G28, K11, R20) The Review of Financial Studies / v 24 n 9 2011 3140 Downloaded from rfs.oxfordjournals.org by guest on September 7, 2011 and Fisher 2011, p.36). The existing evidence seems to support this view, as deficiency judgments are rarely observed. We show that the above viewpoint is false. First, in many US states, residential mortgages are recourse loans. In recourse states, the lender may be able to collect on debt not covered by the proceedings from a foreclosure sale by obtaining a deficiency judgment. Second, the effect of recourse is not reflected in the frequency of deficiency judgments but, instead, in the way the threat of recourse alters borrower behavior. We use loan-level data to empirically investigate the importance of recourse for default behavior. We find that recourse decreases borrowers’ sensitivity to negative equity, i.e., recourse deters some borrowers with negative equity from defaulting. The effect is substantial: On average, the monthly probability of default is 1.32 times higher when there is no threat of recourse than when such a threat is present, when measured at the average value of negative equity for defaulted loans. Ceteris paribus, it takes 8.6% more negative equity to make the probability of default in a recourse situation the same as the probability of default in a non-recourse situation. Importantly, recourse affects default only by lowering borrowers’ sensitivity to negative equity. Unconditionally, there is no difference between the default rates in recourse and non-recourse states. Furthermore, even after controlling for loan and borrower characteristics and including only a dummy for whether the state is recourse or non-recourse in levels, we find that recourse does not have a statistically significant effect on the default rate. We find that recourse has a significant effect on default only when we interact recourse with the default option value. However, there are situations in which recourse does not affect borrower behavior. The effect of recourse is significant only for properties appraised at relatively high values. In particular, we find that, for properties appraised at less than $200,000 (at origination, in real 2005 terms), there is no difference between the probability of default in recourse and non-recourse states. Conversely, for homes appraised at $500,000 to $750,000, at the mean value of the default option at the time of default, borrowers in non-recourse states are more than twice as likely to default as borrowers in recourse states. We also find that allowing the lender recourse increases the likelihood of default by a more lender-friendly method, such as a deed in lieu of foreclosure. This is probably because lenders in recourse states have better bargaining positions. Furthermore, we find that a larger fraction of defaults in non-recourse states are likely to be strategic to the extent that borrowers who default in nonrecourse states are less likely to resume payments following the delinquency that precedes the foreclosure. We also investigate differences in mortgage rates between recourse and nonrecourse states. To the extent that borrowers in recourse states are less likely to default in response to negative equity and, if they do default, are more likely to default in a lender-friendly way, lenders are likely to face smaller losses from Recourse and Residential Mortgage Default: Evidence from US States 1 Pence (2006) examines differences in average loan size in census tracts that span two states and finds that the average loan size is smaller in states with more defaulter-friendly foreclosure laws. 3141 Downloaded from rfs.oxfordjournals.org by guest on September 7, 2011 default in recourse states. Thus, one might expect interest rates to be lower in recourse states. However, we find evidence that loans are more expensive in recourse states. This is a surprising result that requires further investigation. Our finding that recourse deters some borrowers from defaulting indicates that a non-negligible portion of US mortgage default is, in fact, strategic rather than involuntary, i.e., when borrowers have no choice but to default because of liquidity constraints. This finding contrasts with the view that mortgage defaults are primarily driven by shocks to the borrower’s ability to pay. For example, based on their analysis of a rich dataset from Massachusetts, Foote, Gerardi, and Willen (2008) conclude that negative equity is not a sufficient condition for default. However, Massachusetts is a recourse state, and data from recourse states alone give an incomplete picture of the role of negative equity in the borrower’s default decision. As our findings show, the borrower’s decision to default in recourse states is substantially less sensitive to negative equity than in non-recourse states. Guiso, Sapienza, and Zingales (2010) and Bhutta, Dokko, and Shan (2010) also find that at least some portion of default is not caused by liquidity constraints. Earlier work by Clauretie (1987), Jones (1993), and Ambrose, Capone, and Deng (2001) has also empirically examined differences in defaults across jurisdictions.1 Clauretie (1987) estimates a linear regression model of aggregate state default rates and finds that whether a state permits a deficiency judgment does not significantly affect the state’s default rate. Jones (1993) looks at evidence from Alberta, which does not permit deficiency judgments, and British Columbia, which does permit such judgments, and finds that defaults in Alberta are more likely to be deliberate, rather than caused by trigger events in the borrower’s life. Ambrose, Capone, and Deng (2001) include a dummy variable for whether a state allows a deficiency judgment in their study of the determinants of mortgage default in a sample of Federal Housing Administration (FHA) loans originated in 1989. However, as the principal of an FHA loan is guaranteed by the FHA, FHA lenders cannot seek a deficiency judgment, so FHA loans may be poorly suited for studies of the effect of recourse on default behavior. Other related papers include Crawford and Rosenblatt (1995), Ambrose, Buttimer, and Capone (1997), and Corbae and Quintin (2010). Crawford and Rosenblatt (1995) find that, conditional upon default occurring, loss severity is greater in non-recourse states. Our empirical results corroborate Crawford and Rosenblatt’s conclusions, as we find that borrowers in recourse states are more likely to default using a lender-friendly method. Ambrose, Buttimer, and Capone (1997) and Corbae and Quintin (2010) theoretically study the effect of deficiency judgments on default and find that recourse deters default. Our empirical results support these theoretical predictions. The Review of Financial Studies / v 24 n 9 2011 The remainder of the article proceeds as follows. The next section describes how lender recourse varies across US states. Section 2 describes our data and variables. Section 3 shows how recourse affects whether default occurs. Section 4 explores how recourse changes the way default occurs. In Section 5, we look at whether recourse is priced. Section 6 presents our conclusions. 1. Foreclosure Law and Default 3142 Downloaded from rfs.oxfordjournals.org by guest on September 7, 2011 1.1 Foreclosure law across the US states Statutes governing the recourse a lender has in the event it forecloses on a property and the proceeds from the foreclosure sale are not sufficient to cover the borrower’s debt vary from state to state. The length of time it takes a lender to foreclose also varies by state. In most states, when a lender must foreclose in a negative equity situation, that lender may obtain a deficiency judgment to cover the difference between the balance owed and the value of the home. In states that permit deficiency judgments, various restrictions often apply. Usually, the lender must credit the borrower’s account for the fair market value of the property rather than the foreclosure sale price. The fair market value restriction is likely present because the lender is often the only bidder at the foreclosure sale (see, for example, Brueggeman and Fisher 2011). In the absence of such a restriction, the lender could profit from a foreclosure by placing an artificially low bid. In addition to lowering the likely recovery from a deficiency judgment, such restrictions sometimes mean that the lender must incur substantially higher legal costs and spend more time pursuing a deficiency. The increase in costs and time depends on state statutes governing the determination of fair market value. In some states, a single appraiser determines fair market value, while in others, such as Minnesota, fair market value must be determined by a jury. Finally, states differ in how easy it is for the borrower to contest the fair market value assigned to the property. In practice, lenders have less recourse in states that require them to go through a lengthy judicial foreclosure process to obtain a deficiency judgment than in those states characterized by a quicker, non-judicial foreclosure process. Furthermore, in some states, such as Idaho and Nebraska, there is a relatively short period in which the lender can file. In other states, substantial personal property or wages are exempt from collection on the deficiency. Finally, in Ohio and Iowa, the lender has a relatively short period in which to collect on the deficiency after the foreclosure sale. In states that allow deficiency judgments, a borrower retains the option to declare bankruptcy and have some or all of the deficiency judgment discharged. As White (1998) reports, prior to the 2005 bankruptcy reform act (Bankruptcy Abuse Prevention and Consumer Protection Act [BAPCA]), most unsecured debts were discharged in bankruptcy regardless of whether the borrower filed under Chapter 7 or Chapter 13. Furthermore, filing for bankruptcy had a low Recourse and Residential Mortgage Default: Evidence from US States 1.2 Types of default In this article, we use the term “default” to refer to a default that ends with the borrower vacating the home. In practice, lenders usually view litigious foreclosure as a last resort when the borrower defaults and will often try to recover a portion of the principal through other means before resorting to foreclosure.3 When lenders do choose to exercise the option to foreclose, they have a strong interest in foreclosing quickly.4 2 Appendix A describes the foreclosure and deficiency judgment procedures in the US states. We use the fore- closure timelines from the National Mortgage Servicer’s Reference Directory (2004) published by the USFN (America’s Mortgage Banking Attorneys). 3 See, for example, Larsen, Carey, and Carey (2007), Brueggeman and Fisher (2008), and Ling and Archer (2008). 4 This view was also prevalent among the foreclosure attorneys with whom we spoke. 3143 Downloaded from rfs.oxfordjournals.org by guest on September 7, 2011 pecuniary cost before the 2005 act, such that the major cost of filing for bankruptcy was the reduced availability of credit. In Chapter 7 filings, deficiency judgments are still completely discharged and, if the Chapter 7 filing is concurrent with a foreclosure, the lender loses the right to a deficiency judgment. In Chapter 13 filings, the lender may pursue a deficiency judgment. Following the 2005 bankruptcy reform, borrowers with incomes above the state median income must usually file under Chapter 13 rather than under Chapter 7, which might make it more difficult to discharge a deficiency judgment for high-income borrowers. A few states explicitly forbid deficiency judgments on most homes (e.g., Arizona and Oregon) or on purchase mortgages. In other states, the restrictions on deficiency judgments are so onerous that it is highly impractical for the lender to pursue a judgment in the vast majority of cases, which makes the state effectively non-recourse. Table 1 summarizes the extent of recourse the lender has in each state and the time it takes the lender to complete the foreclosure process if the borrower does not contest the foreclosure. We classify Alaska, Arizona, California, Iowa, Minnesota, Montana, North Carolina (purchase mortgages), North Dakota, Oregon, Washington, and Wisconsin as nonrecourse states.2 Our classification of states is similar to that of the USFN (America’s Mortgage Banking Attorneys). The states we classify as non-recourse are the same as those for which the USFN (2004, 5-5 - 5-7) indicates that a deficiency judgment is either not available or highly impractical. However, we classify purchase mortgages in North Carolina as non-recourse because state law prohibits deficiency judgments on purchase mortgages and we treat South Dakota as a recourse state. We typically spoke with at least one foreclosure attorney in every state where the amount of recourse in practice was unclear or the statutes were difficult to understand, which assisted in our determination of the recourse classification. 49–74 108–111 115 90 120 173 160 235 48 200–300 150 48 195 320 150 345 266 180 230 198 209 269 270 46 75 360** 270–280*** 61–65 163 90 NJ NJ NJ NJ NJ NJ J, strict J, by decree of sale NJ J J NJ NJ J NJ J J J J J J, executory process J, non-executory J J J NJ NJ NJ NJ NJ Alabama Alaska Arizona Arkansas California Colorado Connecticut Connecticut DC Delaware Florida Georgia Hawaii Hawaii Idaho Illinois Indiana Iowa Kansas Kentucky Louisiana Louisiana Maine Maryland Massachusetts Michigan Minnesota Missouri Montana Mississippi Recourse Recourse Recourse Recourse Non-Recourse Recourse Non-Recourse Recourse Recourse Recourse Recourse Non-Recourse Recourse Recourse Recourse Recourse Recourse Recourse Recourse Recourse Recourse Non-Recourse Non-Recourse Recourse Non-Recourse Recourse Recourse Recourse Classification Nebraska Nebraska Nevada New Hampshire New Jersey New Mexico New York (NYC) New York (Outside NYC) New York (Outside NYC) North Carolina Purchase Mortgages North Carolina Other Mortgages North Dakota Ohio Oklahoma Oregon Pennsylvania Rhode Island South Carolina South Dakota Tennessee Texas Texas Utah Vermont Virginia Washington West Virginia Wisconsin Wyoming State NJ J NJ NJ J J J J NJ NJ NJ J J NJ NJ J NJ J J NJ NJ J NJ J NJ NJ NJ J NJ Judicial or Non -Judicial Foreclosure 121 176 116 75 295 225 445 299 355 120 120 150 217 201 160 300 74 180 340 50–55 35–60 80–180 139 275 60 140–150 120 315 180 Optimum Timeline* Recourse Recourse Recourse Non-Recourse Recourse Non-Recourse Recourse Non-Recourse Recourse Non-Recourse Recourse Recourse Non-Recourse Recourse Recourse Recourse Recourse Recourse Recourse Recourse Recourse Recourse Recourse Recourse Recourse Recourse Classification *These are optimum timelines from the National Mortgage Servicer’s Reference Directory, 21st edition (2004). The optimum timelines assume no delays and are based on uncontested foreclosure actions. **The non-judicial foreclosure optimally takes 60 days. However, after that, the redemption period begins to run, typically for six months. Estimated time for completion for uncontested foreclosure without eviction action is 12 months. ***The sale in non-judicial foreclosure can generally be held within 90 days. However, there are substantial redemption rights in Minnesota. Thus, including the redemption period, the optimum timeframe for non-judicial foreclosure is 270–280 days. Optimum Timeline* Judicial or Non -Judicial Foreclosure Downloaded from rfs.oxfordjournals.org by guest on September 7, 2011 3144 State Table 1 State foreclosure laws The Review of Financial Studies / v 24 n 9 2011 Recourse and Residential Mortgage Default: Evidence from US States 3145 Downloaded from rfs.oxfordjournals.org by guest on September 7, 2011 Lenders prefer to avoid foreclosures, especially contested foreclosures, for several reasons. First, properties depreciate substantially when the borrower is in default. The depreciation rate is higher when a property is in default because the borrower has no incentive to adequately maintain the property and may deliberately accelerate the property’s depreciation. Second, properties usually sell at depressed values in foreclosure sales. Third, by forcibly removing a borrower from his or her home, lenders may incur negative publicity and reputation costs among other prospective borrowers. Campbell, Giglio, and Pathak (forthcoming) find that a foreclosure reduces the value of the home by approximately 27%. There are at least three lender-friendly ways for a borrower to default: a short sale, a voluntary conveyance, or a simple agreement not to contest the foreclosure. In a short sale, the borrower finds a buyer for the property who pays a purchase price that is less than the full balance of the debt owed. The lender agrees to waive his right to a deficiency in exchange for the borrower selling the property and remitting the proceeds to the lender. Occasionally, the lender may only agree to waive his right to a deficiency if the borrower agrees to give the lender a lump-sum payment in addition to the sale proceedings. In a voluntary conveyance, the borrower hands over the deed to the property to the lender. In the most common form of voluntary conveyance—a deed in lieu—the lender forgives the debt owed in exchange for the deed. In addition to eliminating the risk of the lender pursuing a deficiency judgment, a deed in lieu may affect a borrower’s future access to credit less severely than a forcible eviction (Larsen, Carey, and Carey 2007). The benefit for the lender in this case is that, in addition to getting the property back quickly, the lender’s legal costs are lower and the acceptance of a deed in lieu of foreclosure “can be beneficial to the lender’s public image and to the public perception of the property” (Ling and Archer 2008, p. 231). However, a voluntary conveyance carries some risks for the lender. First, if the borrower declares bankruptcy within one year of a deed in lieu, the court may declare the conveyance improper. In such a case, the lender’s claim becomes an unsecured claim on the borrower’s assets and, in the case of a Chapter 13 filing, on the borrower’s future income, which will generally give the lender a worse payoff. Second, a voluntary conveyance does not cut off any subordinate liens on the property as a foreclosure does. Finally, a borrower may simply agree to what is known as a “friendly foreclosure.” In a friendly foreclosure, the borrower agrees to not contest the foreclosure, and to submit to the jurisdiction of the court in terms of leaving the property and cooperating with the lender. The main benefit of this option is that the lender gets the property back faster than in a contested foreclosure. This option takes more time than a voluntary conveyance but is less time-consuming than a standard foreclosure (Brueggeman and Fisher 2011). A friendly foreclosure may be preferable to the lender because it cuts off any subordinate interests that may exist on the property and protects the lender if the borrower The Review of Financial Studies / v 24 n 9 2011 subsequently declares bankruptcy (Ling and Archer 2008). For the borrower, the benefits of a friendly foreclosure relative to a more standard foreclosure are similar to those arising from a short sale or a deed in lieu: The lender usually agrees to waive his or her right to a deficiency judgment. Subsequent to a voluntary conveyance, the property becomes real estate owned (REO), i.e., the lender owns the property. A property can also become REO subsequent to a foreclosure sale if the lender acquires the property by virtue of being the only bidder. 2. Data The study uses loan-level data obtained from LPS Applied Analytics, Inc. The data contain information on prime and non-prime private securitized loans, portfolio loans, and GSE loans on a monthly basis. Appendix B provides details about the variables by LPS codes. 2.1.1 Default. We define a loan as defaulted if it is terminated in one of the following ways: by REO sale, by short sale, by payoff out of foreclosure, by payoff out of bankruptcy and serious delinquency, or by liquidation to termination. In the analysis of the probability of default, the dependent variable takes a value of 1 in the month the loan defaults. We drop all observations on defaulted loans subsequent to the default month. Consequently, the dependent variable takes a value of 0 for observations in months prior to default for defaulted loans and for all observations on loans that do not default. 2.1.2 Default type. In the analysis of whether recourse changes the way in which default occurs, we consider only defaulted loans. We divide defaults into two categories: default by foreclosure and default by a lender-friendly method, i.e., a short sale or a deed in lieu. We define a default as lender-friendly if the loan passes directly to an REO loan or a short sale. We define a default as a foreclosure if the lender receives a payoff out of bankruptcy or serious delinquency. Such a default is akin to a contested foreclosure process, as the borrower most likely declares bankruptcy to halt foreclosure proceedings. The default type variable takes a value of 1 if the loan defaulted via a foreclosure and 0 otherwise. 2.1.3 Default option variables. We define the value of the default option as the probability that the borrower has negative equity in the house, as in Deng, Quigley, and Van Order (2000) and Ambrose, Capone, and Deng (2001). As we know the balance owed on the loan, we only need to infer the distribution 3146 Downloaded from rfs.oxfordjournals.org by guest on September 7, 2011 2.1 Variable definitions Recourse and Residential Mortgage Default: Evidence from US States of individual house prices. The value of equity to market value ki months after loan i’s origination is E i,t,ki = Mi,t,ki − L i,t,ki , Mi,t,ki where Mi,t,ki is the market value of the property purchased at time t − ki and L i,t,ki is the present value of the remaining loan balance. The market value of the property is H P Ii,t Mi,t,ki = Ci,t−ki , H P Ii,t−ki i σ H2 P Ii,k i is the where Φ(∙) is the cumulative standard normal distribution and variance of individual house prices in the state in which the property associated with mortgage i is located. As in Deng, Quigley, and Van Order (2000), we include the default option linearly and squared. 2.1.4 Prepay option variables. As a proxy for the prepayment option, we use the spread between the current market mortgage rate, rt , and the mortgage rate on the contract, r0 . Our decision to use indicator variables instead of a continuous variable is based on the results presented by Kau, Keenan, and Kim (1994), which show that the spread affects default rates in a nonlinear fashion. Following Ambrose, Capone, and Deng (2001), we define the following dummy variables: Rate1 = 1 if r0 +2% ≤ rt , and 0 otherwise; Rate2 = 1 if r0 +1% ≤ rt < r0 +2% and 0 otherwise; Rate3 = 1 if r0 −1% ≤ rt < r0 +1% and 0 otherwise; Rate4 = 1 if r0 −2% ≤ rt < r0 −1% and 0 otherwise; and Rate5 = 1 if rt < r0 −2% and 0 otherwise, where rt and r0 are in percentages. 5 To calculate the standard deviation H P I /H P I i,t i,t−ki , σ H P Ii,k , we use the volatility parameters A and B i provided by OFHEO as follows: σ H P Ii,k = i q Aki + Bki2 . See Calhoun (1996) for the technical description of OFHEO index. 3147 Downloaded from rfs.oxfordjournals.org by guest on September 7, 2011 where Ci,t−ki is the cost of the property at the time of a purchase, HPI i,t is house price index in the state where the property associated with mortgage i is located, and H P Ii,t /H P Ii,t−ki follows a log-normal distribution (see Case and Shiller 1987 and Deng, Quigley, and Van Order 2000 for details). The mean and variance of H P Ii,t /H P Ii,t−ki are obtained by using the data available from the Office of Federal Housing Enterprise Oversight (OFHEO).5 The value of the default option for mortgage i ki months after origination is the probability that equity is negative:   − ln L M ln i,k i,k i  qi De f ault O ption i,ki = Pr E i,t,ki < 0 = Φ  , σ H2 P Ii,k The Review of Financial Studies / v 24 n 9 2011 2.1.5 Foreclosure timing and recourse variables. We include the time it takes to complete an uncontested foreclosure in the state in which the property is located. Table 1 presents our benchmark recourse classification of states and the foreclosure timelines. We classify purchase mortgages in North Carolina as non-recourse mortgages and other mortgages on property located in North Carolina as recourse mortgages. 2.1.6 Trigger events. We control for trigger events by including the contemporaneous state divorce rate and the state unemployment rate. We use lagged monthly seasonally unadjusted unemployment rates from the Bureau of Labor Statistics.6 2.2 Sample description We use information on loans originated between August 1997 and December 2008. August 1997 is the first month in which the FICO score is available in the data. We restrict our analysis to first mortgages with constant principal and interest, ARMs, or graduated payment mortgages on single-family residences, townhouses, or condos. We drop all FHA and VA loans because deficiency judgments are prohibited on FHA loans and strongly discouraged on VA loans (Larsen, Carey, and Carey 2007). We also drop loans with private mortgage insurance. We then draw a 10% random sample from the LPS database. Our 6 We do not use seasonally adjusted unemployment rates, as there may be a seasonal pattern to defaults due to seasonal economic conditions. 3148 Downloaded from rfs.oxfordjournals.org by guest on September 7, 2011 2.1.7 Loan level variables and borrower characteristics. Additional variables that we use in the empirical analysis are the age of the loan (in months), the loan-to-value (LTV) at origination, an indicator variable that takes a value of 1 if the loan is interest only at origination, an indicator variable that takes a value of 1 if the loan is an adjustable-rate mortgage (ARM), an indicator variable that takes a value of 1 if the loan is a jumbo loan, an indicator variable that takes a value of 1 if the loan is not a purchase mortgage, and the borrower’s FICO score at origination. We convert nominal appraisal amounts at origination into real 2005 dollars by deflating using the CPI excluding shelter. As a mortgage with an LTV of 80% at origination may indicate a higher likelihood of a second mortgage being present, we include a dummy variable that takes a value of 1 if the LTV is exactly equal to 80%. We also include interactions of this variable with the default option value and its square because if an LTV of 80% makes it more likely that the property has a second mortgage, the default option value is higher for these mortgages, such that it may have a stronger effect on the probability of default. See Foote, Gerardi, Goette, and Willen (2009) for empirical evidence that an LTV of exactly 80% increases the risk of default. Recourse and Residential Mortgage Default: Evidence from US States restrictions imply that we have 85,888,286 loan-month observations. In total, our sample includes 2,922,196 loans and 43,353 defaults. 67% of our observations are on recourse mortgages. On average, there is a 1% probability that a homeowner in our sample has negative equity. 7% of our observations are on mortgages that are interest only at origination, while 20% of our observations are on adjustable-rate mortgages. The unconditional default rates are similar in recourse and non-recourse states: In both recourse and non-recourse states, about 1.5% of loans in our sample terminate through a foreclosure, a deed in lieu, or a short sale. Borrowers in recourse states have slightly lower average FICO scores, slightly higher LTVs at origination, and somewhat lower appraisal amounts, and face lower risks of divorce or unemployment. Borrowers in non-recourse states are more likely to have an interestonly mortgage, an adjustable-rate mortgage, and higher default option values. The fraction of mortgages with an LTV of exactly 80% is the same across recourse and non-recourse states. Table 2 provides a summary of the sample. We use a probit as our benchmark model to study the effect of recourse on whether a borrower defaults. We assume that the borrower defaults if an unobserved variable x, x = Xβ + ε, falls below 0, where ε ∼N(0,1). X is a vector of variables that controls for the borrower’s prepay and default options, other loan-level characteristics, and trigger events. The first column of Table 3 presents the results without recourse variables. These results illustrate the effect of the prepay and default options, trigger events, and loan-level characteristics on the default when we do not control for recourse. All of the coefficients have the expected sign. An interest-only loan, an ARM, or a purchase mortgage raises the probability of default. Borrowers with higher FICO scores at origination are less likely to default, while loans with a high LTV at origination are more likely to default. Finally, younger loans are much more likely to default than older loans. The divorce rate has the expected sign but is significant at only the 10% level when we cluster the standard errors, probably because there is relatively little variation across time in a state’s divorce rate. The unemployment rate has the expected sign but becomes insignificant when the standard errors are clustered. Unconditionally, we found no difference in the default rate between recourse and non-recourse states (see Table 2). In column 2 of Table 3, we examine whether the lack of a difference in default rates across recourse regimes is due to differences in loan characteristics, default option values, or prepay option values. The recourse dummy variable takes a value of 1 if the mortgaged property is located in a state with a provision for recourse and 0 otherwise. The coefficient on recourse is statistically insignificant at the 10% level. These results suggest that states that allow recourse for lenders do not have fewer defaults, contrary to Feldstein’s (2008) hypothesis. However, the lack of significance of 3149 Downloaded from rfs.oxfordjournals.org by guest on September 7, 2011 3. The Impact of Recourse on Default 0.47 0.063 0.121 0.360 0.271 0.177 0.87 1.14 61 0.250 0.287 0.397 16 0.87 0.477 3.25 0.35 303,093 Std. Dev. 0 0 0 0 0 0 2.60 3.30 611 0 0 0 33 1.39 0 2 0 82,000 5th Percentile 1 0.021 0 1 1 0 5.10 7.00 799 1 1 1 80 4.22 1 12 1 750,000 95th Percentile 1 0.008 0.015 0.148 0.083 0.038 3.72 4.90 720 0.053 0.061 0.172 68 3.11 0.632 6.58 0.14 275,149 1,924,773 27,927 Mean Recourse Loans 0 0.015 0.015 0.162 0.074 0.022 4.02 5.40 727 0.096 0.148 0.246 64 3.05 0.684 5.74 0.14 391,624 997,423 15,426 Mean Non-Recourse Loans Recourse is a dummy variable that takes a value of 1 if the property is in a recourse state, 0 otherwise. For North Carolina, recourse takes a value of 1 if the loan is not a purchase mortgage, 0 otherwise. The rate variables control for the difference between the current mortgage rate and the contract rate. Purpose Type Dummy takes a value of 1 if the loan is not a purchase mortgage, 0 otherwise. LTV80 Dummy takes a value of 1 if the loan-to-value at origination is 80%. 0.67 0.010 0.015 0.153 0.080 0.033 3.82 5.07 723 0.067 0.090 0.196 67 3.09 0.649 6.30 0.14 308,043 2,922,196 43,353 Mean Downloaded from rfs.oxfordjournals.org by guest on September 7, 2011 3150 Recourse Default Option (Probability of Negative Equity) Rate 1 Rate 2 Rate 4 Rate 5 Divorce Rate Lagged Unemployment Rate Fico Score at Origination Interest Only (at Origination) Dummy Jumbo Dummy ARM Dummy LTV Ratio at Origination Natural Log of Loan Age Purpose Type Dummy Foreclosure Timing (in months) LTV 80 Dummy Appraisal Amount (at Origination) Number of Loans Number of Defaults Table 2 Summary statistics The Review of Financial Studies / v 24 n 9 2011 Recourse and Residential Mortgage Default: Evidence from US States Table 3 Recourse and the probability of default No Recourse Dummies Default Option Default Option Squared Recourse Default Option * Recourse Default Option Sq. * Recourse LTV80 Rate 2 Rate 4 Rate 5 Divorce Rate Lagged Unemp Rate Fico Score at Origination Interest Only Dummy Jumbo Dummy ARM Dummy LTV Ratio at Origination Ln Loan Age Purpose Type Dummy Constant % Defaults Log ps. likelihood Pseudo R−squared Number of obs. Recourse in Interactions (1) (2) (3) 1.06 (0.25) −1.18 (0.22) − 1.97 (0.19) −2.22 (0.21) − − 1.06 (0.25) −1.21 (0.23) −0.0620 (0.0409) − − − 0.13 (0.02) 0.98 (0.30) −1.81 (0.50) −0.359 (0.027) −0.311 (0.035) 0.326 (0.011) 0.444 (0.014) 0.023 (0.014) 0.008 (0.018) −0.263 (0.014) 0.195 (0.023) 0.046 (0.020) 0.313 (0.014) 0.009 (0.001) 0.078 (0.005) −0.097 (0.020) −2.75 (0.13) 0.050% −311,161 16.46% 85,888,286 0.13 (0.02) 0.93 (0.28) −1.75 (0.45) −0.354 (0.025) −0.310 (0.035) 0.328 (0.011) 0.450 (0.015) 0.020 (0.014) 0.005 (0.022) −0.265 (0.015) 0.191 (0.022) 0.031 (0.016) 0.308 (0.011) 0.009 (0.001) 0.081 (0.007) −0.098 (0.021) −2.68 (0.16) 0.050% −311,001 16.50% 85,888,286 −1.52 (0.35) 1.57 (0.43) 0.13 (0.02) 0.57 (0.19) −1.40 (0.30) −0.350 (0.025) −0.306 (0.033) 0.327 (0.011) 0.451 (0.015) 0.023 (0.014) 0.007 (0.020) −0.264 (0.015) 0.187 (0.022) 0.027 (0.017) 0.310 (0.012) 0.010 (0.001) 0.077 (0.005) −0.099 (0.021) −2.78 (0.13) 0.050% −310,662 16.59% 85,888,286 Recourse in Levels and Interactions (4) 1.89 (0.13) −2.13 (0.15) −0.0237 (0.0379) −1.38 (0.21) 1.42 (0.29) 0.13 (0.02) 0.58 (0.19) −1.41 (0.30) −0.349 (0.025) −0.306 (0.033) 0.328 (0.011) 0.452 (0.016) 0.022 (0.014) 0.006 (0.023) −0.265 (0.015) 0.186 (0.021) 0.023 (0.015) 0.309 (0.010) 0.010 (0.001) 0.078 (0.006) −0.099 (0.022) −2.75 (0.16) 0.050% −310,643 16.60% 85,888,286 Recourse in Interactions with State Dummies (5) 1.99 (0.20) −2.17 (0.21) − −1.65 (0.30) 1.82 (0.37) 0.13 (0.02) 0.58 (0.14) −1.31 (0.26) −0.359 (0.024) −0.317 (0.032) 0.337 (0.008) 0.465 (0.013) − −0.054 (0.015) −0.264 (0.015) 0.187 (0.021) 0.021 (0.014) 0.289 (0.009) 0.010 (0.001) 0.087 (0.007) −0.113 (0.020) −2.28 (0.12) 0.050% −308,237 17.24% 85,888,286 The dependent variable in the probit is a binary variable that takes a value of 1 if the loan defaults in that month, 0 otherwise. Default Option is the probability that the borrower has negative home equity. LTV80 takes a value of 1 if the loan has an LTV of exactly 80%, 0 otherwise. The rate variables control for the difference between the current mortgage rate and the contract rate. Purpose Type Dummy takes a value of 1 if the loan is not a purchase mortgage, 0 otherwise. Standard errors are in parentheses. The coefficients and standard errors for Fico Score at Origination show the effect of a 100-point increase in the FICO score. Standard errors are clustered by state. Coefficients in bold are significant at the 5% level. % Defaults is the percentage of observations that are defaults. 3151 Downloaded from rfs.oxfordjournals.org by guest on September 7, 2011 LTV80 * Default Option LTV80 * Default Option Sq. Rate 1 Recourse in Levels The Review of Financial Studies / v 24 n 9 2011 7 All of our results are similar when we do not include the LTV80 variable and its interactions. 8 We drop the divorce rate in this specification as our divorce rate data are only available at an annual frequency. Also, for some states, we only have a few divorce rate observations over the entire sample, such that there is little variation remaining in the divorce rate after we control for state-specific effects. 3152 Downloaded from rfs.oxfordjournals.org by guest on September 7, 2011 recourse in levels may be due to other differences across states in either the legal or economic environments. Recourse directly affects a borrower’s payoff from defaulting. Different payoffs from the default decision in recourse and non-recourse states may lead to different threshold values of the default option at which the borrower defaults in recourse and non-recourse states. Thus, the appropriate specification with which to estimate the impact of recourse on the probability of default models recourse as an interaction term between the value of the default option and the recourse indicator variable. Column 3 of Table 3 contains the main finding of this article. The coefficient on the interaction term between recourse and negative equity is negative and statistically significant. The coefficient on the interaction term between recourse and the square of the probability of negative equity is positive and statistically significant. The negative coefficient on the default option value indicates that recourse decreases the impact of negative equity on the probability of default. The positive coefficient on the squared term indicates that the effect decreases as the default option value increases. Given this nonlinear effect of default option value on the probability of default, the effect of recourse depends on the particular value of the default option. The coefficient on the interaction between the default option value and the dummy for an LTV of exactly 80% is significant and positive, suggesting that properties with first mortgages that have an LTV of exactly 80% are more likely to have second mortgages attached to them (see Foote, Gerardi, Goette, and Willen 2009), and thus are more sensitive to negative equity. The inclusion of this term is important for ensuring that the results are not driven by the fact that there may be more second mortgages in non-recourse states than in recourse states.7 In column 4 of Table 3, we include recourse in levels as well as in interactions with the default option value. The coefficient on the interaction between the default option value and recourse is again negative and highly significant. It is only slightly smaller in magnitude than in the specification in column 3, where we include recourse only in interactions. In column 4, the coefficient on recourse in levels falls substantially and is far from statistically significant. The results corroborate the hypothesis that recourse affects the payoff from defaulting; recourse only affects a state’s default rate by decreasing borrower sensitivity to the put option. In column 5 of Table 3, we explicitly incorporate the sorts of state-specific factors that recourse in levels likely captures by including state dummy variables.8 This enables us to ensure that our results regarding the effect of recourse Recourse and Residential Mortgage Default: Evidence from US States 3.1 Alternative specifications In columns 2 through 4 of Table 5, we present the results for two alternative specifications. In column 2, we include the prepay option—the difference between the contract rate and current mortgage rates—in interactions with the probability of negative equity, as in Ambrose, Capone, and Deng (2001). The results are similar to our benchmark specification, although the log-likelihood is somewhat higher when rates are included in interactions, which suggests that the inclusion of rates in levels fits the data better. 3.1.1 Foreclosure timing. In column 3 of Table 5, we show the effect of the lengthiness of the uncontested foreclosure process, as stated in USFN (2004), on the probability of default. In column 3, we include the length (in months) of the uncontested foreclosure process for the state in which the property is located. When we do not cluster the standard errors, states with longer foreclosure processes appear to experience more defaults. However, the effect becomes insignificant when we cluster the standard errors by state. We also do not find that the lengthiness of the foreclosure process significantly affects default in other specifications in which we include the interaction of the foreclosure timeframe with recourse. We obtain similar results with a specification in which we include foreclosure timing by using a dummy variable that takes a 3153 Downloaded from rfs.oxfordjournals.org by guest on September 7, 2011 on borrower sensitivity to the default option are not due to state-specific factors. When we control for state-specific fixed effects, the results on the effect of recourse carry through: The coefficient on the interaction between recourse and the default option value is statistically significant, negative, and slightly larger in magnitude than in the benchmark specification. Thus, our results regarding the deterrent effect of recourse are not driven by unobserved differences between recourse and non-recourse states. To gauge the magnitude of the deterrent effect of recourse, we evaluate the probability of default in recourse and non-recourse states at different values of the default option. Table 4 contains the estimates of the probabilities. Columns 1 through 4 show the probabilities at the means of the continuous variables and the modes of the dummy variables. At the mean of the default option for all observations, the probability of default is 6% higher in non-recourse states than in recourse states. At the 90th percentile of the value of the default option for all observations, the probability of default in non-recourse states is 2% higher. This difference increases to 13% at the 95th percentile. The results in Table 4 indicate that recourse has a deterrent effect on default at high values of the default option value, which are precisely the values associated with default. At the mean of the default option at the time of default, borrowers in non-recourse states are 32% more likely to default than borrowers in recourse states. Thus, the data allow us to reject the hypothesis that recourse has no effect on default. The Review of Financial Studies / v 24 n 9 2011 Table 4 Estimated default probabilities in recourse and non-recourse states At Time of Default At Mean of Default (1) 6.08% 0.1801% 0.1361% 132% Default Option Value Non-Recourse Def. Prob. Recourse Def. Prob. Ratio NR/R 6.09% 0.1750% 0.1263% 139% Default Option Value Non-Recourse Def. Prob. Recourse Def. Prob. Ratio NR/R 8.84% 0.2141% 0.1185% 181% Default Option Value Non-Recourse Def. Prob. Recourse Def. Prob. Ratio NR/R 10.12% 0.1721% 0.0756% 228% Default Option Value Non-Recourse Def. Prob. Recourse Def. Prob. Ratio NR/R 8.54% 0.0818% 0.0512% 160% Default Option Value Non-Recourse Def. Prob. Recourse Def. Prob. Ratio NR/R 6.41% 0.3306% 0.2283% 145% Default Option Value Non-Recourse Def. Prob. Recourse Def. Prob. Ratio NR/R 10.16% 0.0807% 0.0558% 144% (2) Benchmark Specification 1.00% 0.29% 0.0088% 0.0083% 0.0083% 0.0082% 106% 102% By Appraisal Amount (Real, 2005$) $200,000 to $300,000 0.84% 0.23% 0.0069% 0.0066% 0.0065% 0.0065% 106% 102% $300,000 to $500,000 1.14% 0.24% 0.0048% 0.0045% 0.0044% 0.0044% 111% 102% $500,000 to $750,000 1.57% 0.32% 0.0043% 0.0039% 0.0036% 0.0037% 120% 104% $750,000 to $1,000,000 1.30% 0.17% 0.0039% 0.0035% 0.0036% 0.0035% 109% 101% By Investor Type Private Securitized 2.21% 2.47% 0.0231% 0.0235% 0.0196% 0.0196% 118% 120% Private Portfolio 2.55% 2.80% 0.0211% 0.0215% 0.0189% 0.0191% 111% 113% At 95th percentile (4) 2.07% 0.0095% 0.0084% 113% 1.28% 0.0072% 0.0066% 109% 2.01% 0.0052% 0.0044% 120% 4.00% 0.0053% 0.0034% 156% 2.55% 0.0043% 0.0037% 119% 11.40% 0.0413% 0.0196% 210% 15.37% 0.0448% 0.0257% 174% The benchmark specification is specification (3) (recourse in interactions) from Table 3. The probabilities are estimated at the modes for dummy variables and at the means for the variables other than the default option value and default option value squared. In column 1, we estimate the probabilities at the modes of dummy variables and at the means of all variables at the time of default for defaulted loans. Ratio is the ratio of the probabilities in non-recourse and recourse states. value of 1 if the state’s uncontested foreclosure process takes more than six months and 0 otherwise. 3.1.2 A finer recourse classification. In our benchmark specification, we define mortgages as either recourse or non-recourse. Our benchmark classification (see Table 1) defines a mortgage as non-recourse if deficiency judgments are either explicitly prohibited or impractical. We also consider a finer 3154 Downloaded from rfs.oxfordjournals.org by guest on September 7, 2011 Default Option Value Non-Recourse Def. Prob. Recourse Def. Prob. Ratio NR/R At Mean All Loans Value of Default Option At 90th percentile (3) Recourse and Residential Mortgage Default: Evidence from US States Table 5 Alternative specifications Benchmark (1) Default Option Default Option Squared Default Option * Recourse Default Option Sq. * Recourse Default Opt. * Limited Recourse Default Opt. Sq. * Limited Recourse LTV80 LTV80 * Default Option Rate 1 Rate 2 Rate 4 Rate 5 Default Option * Rate 1 − 0.13 (0.02) 0.57 (0.19) −1.40 (0.30) −0.350 (0.025) −0.306 (0.033) 0.327 (0.011) 0.451 (0.015) − Default Option * Rate 5 − Jumbo Dummy ARM Dummy LTV Ratio at Origination Ln Loan Age Purpose Type Dummy Foreclosure Timing Constant % Defaults − − −2.35 (0.17) 0.050% − Interest Only Dummy − −2.78 (0.13) 0.050% Default Option * Rate 4 Fico Score at Origination 0.12 (0.02) 0.87 (0.20) −1.74 (0.30) − 0.023 (0.014) 0.007 (0.020) −0.264 (0.015) 0.187 (0.022) 0.027 (0.017) 0.310 (0.012) 0.010 (0.001) 0.077 (0.005) −0.099 (0.021) − − Lagged Unemp Rate − −2.418 (1.121) −2.210 (0.492) 0.796 (0.053) 0.906 (0.093) 0.024 0.013 0.015 (0.019) −0.348 (0.021) 0.111 (0.023) −0.019 (0.018) 0.375 (0.008) 0.013 (0.001) 0.065 (0.006) −0.099 (0.021) − Default Option * Rate 2 Divorce Rate (2) 1.97 (0.22) −2.81 (0.27) −1.51 (0.35) 1.33 (0.47) − Foreclosure Timing (3) 1.98 (0.17) −2.23 (0.20) −1.55 (0.31) 1.60 (0.40) − Finer Recourse Classification (4) 0.13 (0.02) 0.57 (0.19) −1.40 (0.30) −0.350 (0.025) −0.306 (0.033) 0.328 (0.010) 0.451 (0.014) − 2.32 (0.21) −2.83 (0.27) −1.86 (0.36) 2.17 (0.46) −0.68 (0.20) 1.19 (0.24) 0.13 (0.02) 0.53 (0.19) −1.35 (0.30) −0.350 (0.025) −0.306 (0.033) 0.327 (0.010) 0.451 (0.015) − − − − − − − 0.025 (0.014) 0.006 (0.017) −0.264 (0.015) 0.188 (0.021) 0.028 (0.017) 0.310 (0.012) 0.010 (0.001) 0.078 (0.005) −0.099 (0.021) 0.003 (0.007) −2.80 (0.16) 0.050% 0.022 (0.015) 0.007 (0.021) −0.264 (0.015) 0.186 (0.022) 0.027 (0.017) 0.310 (0.012) 0.010 (0.001) 0.077 (0.005) −0.096 (0.021) − − −2.78 (0.13) 0.050% (continued) 3155 Downloaded from rfs.oxfordjournals.org by guest on September 7, 2011 LTV80 * Default Option Sq. 1.97 (0.19) −2.22 (0.21) −1.52 (0.35) 1.57 (0.43) − Rates in Interactions The Review of Financial Studies / v 24 n 9 2011 Table 5 Continued Log ps. likelihood Pseudo R−squared Number of obs. Benchmark Rates in Interactions Foreclosure Timing (1) (2) (3) −310,662 16.59% 85,888,286 −318,205 14.57% 85,888,286 −310,649 16.59% 85,888,286 Finer Recourse Classification (4) −310,621 16.60% 85,888,286 The dependent variable in the probit is a binary variable that takes a value of 1 if the loan defaults in that month, 0 otherwise. Default Option is the probability that the borrower has negative home equity. LTV80 takes a value of 1 if the loan has an LTV of exactly 80%, 0 otherwise. The rate variables control for the difference between the current mortgage rate and the contract rate. Purpose Type Dummy takes a value of 1 if the loan is not a purchase mortgage, 0 otherwise. Standard errors are in parentheses. The coefficients and standard errors for FICO Score at Origination show the effect of a 100-point increase in the FICO score. Standard errors are clustered by state. Coefficients in bold are significant at the 5% level. % Defaults is the percentage of observations that are defaults. 3.2 Robustness We conduct several additional robustness exercises. First, we repeat our analysis using only data on mortgages originated from 2005 onward, as there is some concern that the data covering these mortgages are of higher quality. In addition, a very large mortgage servicer enters the database in 2005. Column 1 of Table 6 presents the results from our benchmark specification using only data on originations from 2005 to the end of our sample period. Column 2 presents the results of the specification that includes state fixed effects using 3156 Downloaded from rfs.oxfordjournals.org by guest on September 7, 2011 classification of non-recourse. In this specification, we differentiate between de jure (i.e., explicitly) non-recourse mortgages and de facto non-recourse (i.e., limited recourse) mortgages. We define mortgages on properties in Arizona, North Dakota, and Oregon, as well as purchase mortgages in California, Montana, and North Carolina, as de jure non-recourse. We define California and Montana non-purchase mortgages, as well as mortgages on properties in Alaska, Iowa, Minnesota, Washington, and Wisconsin, as de facto non-recourse. In column 4 of Table 5, we present the results from the specification in which we use this finer recourse classification: recourse (the same mortgages as in the benchmark specification), de facto non-recourse, and de jure nonrecourse. The omitted category is de jure non-recourse; limited recourse is a dummy variable that takes a value of 1 if the mortgage is de facto non-recourse and 0 otherwise. The coefficient on recourse remains significantly negative and is slightly larger in magnitude than in our benchmark specification. The coefficient on limited recourse is significantly negative but much smaller in magnitude than the coefficient on recourse. Thus, default is more likely if the mortgage is de jure non-recourse than if it is de facto non-recourse. However, default is more likely if the mortgage is de facto non-recourse than if it is recourse. Recourse and Residential Mortgage Default: Evidence from US States Table 6 Robustness 2005–2008 Sample Coefficients State Dummies Benchmark Default Option Default Option Squared Default Option * Recourse Default Option Sq. * Recourse LTV80 LTV80 * Default Option LTV80 * Default Option Sq. Rate 1 Rate 4 Rate 5 Divorce Rate Lagged Unemp Rate Fico Score at Origination Interest Only Dummy Jumbo Dummy ARM Dummy LTV Ratio at Origination Ln Loan Age Purpose Type Dummy Constant % Defaults Log ps. likelihood Pseudo R-squared Number of obs. (2) 1.27 (0.10) −1.46 (0.12) −1.53 (0.29) 1.79 (0.34) 0.14 (0.02) 0.08 (0.14) −0.47 (0.25) −0.016 (0.05) −0.426 (0.034) 0.338 (0.011) 0.460 (0.013) − −0.115 (0.008) −0.267 (0.018) 0.131 (0.018) 0.013 (0.030) 0.216 (0.018) 0.011 (0.002) 0.217 (0.026) −0.173 (0.019) −2.14 (0.12) 0.105% −178,135 16.3% 25,828,688 1.47 (0.19) −1.80 (0.18) −1.60 (0.36) 1.70 (0.43) 0.12 (0.02) 0.42 (0.23) −1.11 (0.36) −0.285 (0.03) −0.275 (0.02) 0.321 (0.01) 0.425 (0.01) 0.031 (0.015) 0.016 (0.019) −0.274 (0.013) 0.145 (0.015) 0.031 (0.017) 0.260 (0.008) 0.010 (0.001) 0.135 (0.020) −0.097 (0.021) −3.07 (0.14) 0.050% −308,656 17.1% 85,888,286 Full Sample Hazard Ratios Competing Competing Hazards, Hazards, Separate Est. Joint Est. (5) (4) 1.057 (0.002) 0.999 (0.000) 0.958 (0.002) 1.000 (0.000) 1.562 (0.020) 1.014 (0.003) 1.000 (0.000) 0.279 (0.023) 0.324 (0.010) 2.857 (0.040) 3.877 (0.070) 1.071 (0.006) 1.100 (0.005) 0.992 (0.000) 1.802 (0.027) 1.074 (0.017) 2.840 (0.039) 1.035 (0.001) 0.224 (0.004) 0.602 (0.007) − −559,643 1.060 (0.003) 0.999 (0.000) 0.958 (0.003) 1.000 (0.000) 1.571 (0.030) 1.008 (0.004) 1.000 (0.000) 0.274 (0.035) 0.302 (0.016) 2.985 (0.066) 4.103 (0.118) 1.069 (0.010) 1.079 (0.008) 0.993 (0.000) 1.794 (0.042) 1.084 (0.026) 2.881 (0.062) 1.032 (0.001) 0.130 (0.001) 0.698 (0.012) − −12,033,972 85,888,286 The dependent variable in the probit (columns 1–3) is a binary variable that takes a value of 1 if the loan defaults in that month, 0 otherwise. Default Option is the probability that the borrower has negative home equity. LTV80 takes a value of 1 if the loan has an LTV of exactly 80%. The rate variables control for the difference between the current mortgage rate and the contract rate. Purpose Type Dummy takes a value of 1 if the loan is not a purchase mortgage, 0 otherwise. Standard errors are in parentheses. The coefficients and standard errors for Fico Score at Origination show the effect of a 100-point increase in the FICO score. Columns 4 and 5 present hazard ratios from the estimated proportional hazard semiparametric models; standard errors are clustered by loan. The results in column 4 are from hazard functions for grouped data estimated separately, treating the competing hazard as censored. The results in column 5 are from the joint estimation of competing hazard functions for grouped data; due to computational constraints, we reduce our sample to a random 40% sample in this specification. Standard errors are clustered by state for the probit models. Coefficients in bold are significant at the 5% level. % Defaults is the % of observations that are defaults. 3157 Downloaded from rfs.oxfordjournals.org by guest on September 7, 2011 Rate 2 (1) 1.53 (0.20) −1.84 (0.18) −1.94 (0.43) 2.18 (0.48) 0.16 (0.02) 0.12 (0.19) −0.75 (0.31) −0.040 (0.05) −0.402 (0.035) 0.333 (0.012) 0.430 (0.012) 0.060 (0.022) −0.005 (0.021) −0.260 (0.016) 0.136 (0.020) 0.070 (0.018) 0.277 (0.012) 0.011 (0.002) 0.161 (0.020) −0.154 (0.017) −3.09 (0.17) 0.105% −180,901 15.0% 25,828,688 Coefficients Year of Origination Dummies (3) The Review of Financial Studies / v 24 n 9 2011 9 The full set of results using only data from 2005 onward is available from the authors. 3158 Downloaded from rfs.oxfordjournals.org by guest on September 7, 2011 only data on originations from 2005 to the end of the sample period. The results are consistent with those we obtain using our benchmark specification.9 Columns 3 through 5 of Table 6 present additional results for the full sample. Column 3 contains the results from a specification that includes dummies for the year of origination. The coefficient on the interaction between the default option value and recourse is similar to that for the benchmark specification (column 3 of Table 3). With 2003 as the omitted category, the coefficients on origination years from 1998 to 2001 are negative and statistically significant, while the coefficients on 2004, 2005, and 2006 are positive and statistically significant. These results provide some evidence that, when controlling for a set of variables used in our benchmark specification, the mortgages originated in the later years of the sample period, particularly from 2004 to 2006, have a higher probability of default than the mortgages originated earlier. In columns 4 and 5 of Table 6, we present proportional competing hazard models (see, for example, Deng, Quigley, and Van Order 2000). In our benchmark specification, we use a probit model and control for time dependence using the time elapsed from loan origination. Thus, our benchmark specification provides an estimate of the probability of the loan defaulting in any particular month. Alternatively, we can estimate the hazard model of the risk of default. Generally, a mortgage can be terminated by default or prepayment, such that a mortgage is subject to two competing hazards. Column 4 of Table 6 contains the results (for default) from fitting models for each termination type separately and treating failures due to a competing type of termination as censored data. The hazard ratio on the interaction between recourse and the default option is less than 1 and highly significant, indicating that recourse reduces the sensitivity of default to negative equity, as we found using our benchmark specification. We also estimate the two hazard functions jointly. To do so, we assume that the two competing hazard functions are additive. Consequently, the hazard of failure by any termination type is the sum of the two competing processes. The observed time of failure is the minimum time of failure of the two competing processes. Thus, at the time of failure, two survival times are observed: one for a process that corresponds to the failure type and another, which is censored, that corresponds to the competing process. We use a proportional hazard model with grouped duration data. To estimate the competing hazards of default and prepayment, we duplicate the data using the method suggested by Lunn and McNeil (1995). The duplicated dataset contains twice as many observations as the original dataset, with each new observation showing a censored observation for a competing termination type. The censored observations are also duplicated, which creates two censored observations—one for each failure type. We then define a variable that identifies two strata: one for prepayment and one for default. The failure indicator Recourse and Residential Mortgage Default: Evidence from US States then reflects failures from a type of termination corresponding to the respective stratum. We estimate the semi-parametric Cox model including a strata indicator as a covariate, as well as interactions of the strata indicator with all covariates, in addition to our benchmark covariates. The inclusion of the strata indicator as a covariate assumes proportional baseline hazards for the two competing types of termination while allowing the effect of covariates on the hazard to differ. Column 5 of Table 6 presents the results for default from the joint estimation of the competing hazards model. The results obtained from jointly estimating competing hazards are similar to those obtained when the hazards are estimated separately. The hazard ratio on the interaction between recourse and the default option is less than 1 and highly statistically significant. We also estimate a version of the model in which the default option enters non-parametrically. The results are similar to those for our benchmark specification and are available from the authors. 3159 Downloaded from rfs.oxfordjournals.org by guest on September 7, 2011 3.3 Results by appraisal amount The deterrent effect of recourse on default depends on the lender’s recovery rate, i.e., the fraction of the deficiency judgment that a lender can actually recover. We proxy for the lender’s recovery rate with the appraised value of the mortgaged property. A higher appraised value is likely to indicate that the borrower has more assets that the lender can use to recover in a deficiency judgment. Additionally, a higher appraisal amount is more likely to be associated with higher income, as the ratio of debt to income is a key ratio in the underwriting process. Higher-income borrowers who declare bankruptcy may also have less chance to have their debt discharged during bankruptcy proceedings. This is particularly true for borrowers considering default after the 2005 bankruptcy reform, which usually requires borrowers with income above the state median to file under Chapter 13 rather than under Chapter 7. This implies that lenders should have better recovery rates with richer borrowers. In fact, our examination of the data from the Survey of Consumer Finance indicates that there is a positive relationship between the median value of the primary residence and financial (non-housing) wealth. In 2007, households in the lowest quintile of financial (non-housing) wealth held homes worth $81,946 on average, while households in the second, third, fourth, and fifth quintiles of nonhousing wealth held homes worth on average $118,367, $154,788, $191,208, and $318,681, respectively (in real 2004 dollars). Table 7 presents the results of estimating our benchmark specification separately for different values of the appraised value (real 2005 dollars) of the mortgaged property at origination. As the results in Table 6 show, recourse does not deter default for all households in the same way. Recourse is a deterrent for default when the appraisal amount exceeds $200,000. The coefficient on the recourse interaction with the default option value and its square Lagged Unemp Divorce Rate Rate 5 Rate 4 Rate 2 LTV80 * Default Option LTV80 * Default Option Sq. Rate 1 Default Option Squared Default Option * Recourse Default Option Sq. * Recourse LTV80 Default Option 1.97 (0.19) –2.22 (0.21) –1.52 (0.35) 1.57 (0.43) 0.13 (0.02) 0.57 (0.19) –1.40 (0.30) –0.350 (0.025) –0.306 (0.033) 0.327 (0.011) 0.451 (0.015) 0.023 (0.014) 0.007 0.99 (0.34) –1.79 (0.61) 0.18 (0.36) –0.20 (0.76) 0.05 (0.01) 2.64 (0.92) –16.11 (5.08) –0.200 (0.098) –0.182 (0.031) 0.233 (0.016) 0.364 (0.025) 0.017 (0.024) 0.046 <$100,000 1.22 (0.36) –1.14 (0.44) –0.61 (0.39) 0.31 (0.48) 0.08 (0.01) 0.86 (0.39) –2.44 (0.66) –0.274 (0.049) –0.227 (0.023) 0.259 (0.014) 0.372 (0.021) −0.015 (0.018) 0.017 $100,000 to $200,000 2.18 (0.20) –2.45 (0.25) –1.76 (0.30) 1.95 (0.34) 0.13 (0.01) 0.75 (0.45) –1.59 (0.97) –0.325 (0.036) –0.314 (0.037) 0.321 (0.018) 0.468 (0.019) 0.025 (0.016) –0.012 $200,000 to $300,000 $500,000 to $750,000 2.25 (0.20) –2.86 (0.27) –2.81 (0.44) 3.66 (0.60) 0.21 (0.01) –0.56 (0.12) 0.39 (0.20) –0.547 (0.069) –0.487 (0.043) 0.399 (0.014) 0.604 (0.031) 0.104 (0.025) –0.026 $300,000 to $500,000 2.27 (0.23) –2.55 (0.25) –2.28 (0.36) 2.44 (0.44) 0.19 (0.01) 0.29 (0.21) –1.14 (0.32) –0.354 (0.046) –0.325 (0.048) 0.387 (0.024) 0.519 (0.025) 0.067 (0.021) –0.024 Downloaded from rfs.oxfordjournals.org by guest on September 7, 2011 3160 All Table 7 Recourse and the probability of default by appraisal amount (real 2005 dollars) 2.24 (0.22) –2.49 (0.29) –1.67 (0.82) 1.22 (1.23) 0.20 (0.02) –1.02 (0.21) 0.85 (0.28) –0.353 (0.080) –0.371 (0.090) 0.515 (0.056) 0.696 (0.068) 0.061 (0.027) –0.040 $750,000 to $1,000,000 (continued) 1.71 (0.33) –2.22 (0.40) 1.26 (1.16) –5.30 (2.65) 0.06 (0.04) −0.04 (0.23) –0.14 (0.33) –0.203 (0.066) –0.282 (0.052) 0.471 (0.030) 0.699 (0.050) 0.019 (0.032) –0.022 > $1,000,000 The Review of Financial Studies / v 24 n 9 2011 <$100,000 (0.021) –0.206 (0.020) 0.010 (0.035) 0.291 (0.014) 0.006 (0.001) 0.039 (0.005) –0.050 (0.016) 0.115% –55,730 11.52% 7,076,266 All (0.020) –0.264 (0.015) 0.187 (0.022) 0.313 (0.014) 0.009 (0.001) 0.078 (0.005) –0.097 (0.020) 0.050% –311,161 16.46% 85,888,286 (0.020) –0.279 (0.013) 0.146 (0.023) 0.325 (0.015) 0.009 (0.000) 0.080 (0.005) –0.084 (0.016) 0.048% –96,082 14.62% 26,973,983 $100,000 to $200,000 (0.018) –0.291 (0.012) 0.206 (0.018) 0.344 (0.018) 0.009 (0.001) 0.096 (0.009) –0.108 (0.022) 0.039% –57,104 17.72% 20,004,481 $200,000 to $300,000 (0.016) –0.291 (0.018) 0.202 (0.011) 0.330 (0.019) 0.013 (0.001) 0.106 (0.014) –0.154 (0.031) 0.046% –63,350 20.45% 19,840,252 $300,000 to $500,000 Downloaded from rfs.oxfordjournals.org by guest on September 7, 2011 (0.020) –0.292 (0.026) 0.231 (0.017) 0.206 (0.016) 0.015 (0.002) 0.101 (0.019) –0.145 (0.023) 0.050% –26,487 20.36% 7,675,467 $500,000 to $750,000 (0.025) –0.266 (0.021) 0.209 (0.029) 0.195 (0.033) 0.012 (0.003) 0.069 (0.019) –0.086 (0.023) 0.034% –5,866 18.29% 2,370,213 $750,000 to $1,000,000 (0.024) –0.280 (0.026) 0.175 (0.051) 0.102 (0.041) 0.013 (0.001) 0.076 (0.020) –0.043 (0.027) 0.027% –4,121 14.55% 1,947,624 > $1,000,000 The dependent variable in the probit is a binary variable that equals 1 if the loan defaults in that month, 0 otherwise. The benchmark specification is specification (3) from Table 3 (recourse in interactions). Default Option is the probability that the borrower has negative home equity. LTV80 takes a value of 1 if the loan has an LTV of exactly 80%, 0 otherwise. The rate variables control for the difference between the current mortgage rate and the contract rate. The results for Fico Score at Origination show the effect of a 100-point increase in the FICO score. Purpose Type Dummy takes a value of 1 if the loan is not a purchase mortgage, 0 otherwise. % Defaults is the percentage of observations that are defaults. Standard errors (clustered by state) are in parentheses. All regressions include a constant. Purpose Type Dummy % Defaults Log ps. likelihood Pseudo R-sq Number of obs. LTV Ratio at Origination Ln Loan Age Rate Fico Score at Origination Interest Only Dummy ARM Dummy Table 7 Continued Recourse and Residential Mortgage Default: Evidence from US States 3161 The Review of Financial Studies / v 24 n 9 2011 3.4 Recourse and lender types Table 8 presents the results from the probit regression estimated separately for loans held by Fannie Mae (FNMA), loans held by Freddie Mac (FHMLC), loans that are privately held and securitized, and loans held in a bank’s portfolio. As seen in Table 8, the coefficient on the interaction of the recourse dummy with the default option value is negative, sizeable, and statistically significant for privately securitized and private portfolio loans. Table 4 presents estimates 3162 Downloaded from rfs.oxfordjournals.org by guest on September 7, 2011 are statistically insignificant when the appraisal amount is $200,000 or less. The coefficient on the interaction of the recourse with a linear default option term is particularly large in the samples with appraisal amounts from $300,000 to $500,000 and from $500,000 to $750,000. For the sample with appraisal amounts of $1,000,000 or higher, the coefficient changes sign and is not statistically significant. The results of the estimation of the probability of default in the samples by appraisal amount indicate that the effect of recourse on the probability of default is mainly driven by borrowers with mortgages on properties appraised at $200,000 or more. To the extent that the appraisal amount at origination proxies for the recovery rate on a deficiency judgment, these results indicate that recourse has a substantial deterrent effect on default in cases with higher recovery rates. Recourse does not have a statistically significant effect when the recovery on a deficiency judgment is likely to be low. To gauge the magnitude of the deterrent effect of recourse on the default probabilities, we present estimates of the probabilities of default in recourse and nonrecourse states in Table 4. At the mean value of the default option at the time of default and for homes appraised at $300,000 to $500,000, borrowers in non-recourse states are 81% more likely to default than borrowers in recourse states. For homes appraised at $500,000 to $750,000, borrowers in non-recourse states are more than twice as likely to default as borrowers in recourse states. For homes appraised at $750,000 to $1,000,000, borrowers in non-recourse states are 60% more likely to default than borrowers in recourse states. Importantly, the size of the deficiency judgment relative to the lender’s fixed cost of filing for a deficiency is likely to be lower for low-value properties than for high-value properties. This lowers the incentive for a lender to file for a deficiency judgment for low-value properties. If the recovery rate is 100%, costs do not matter because they are recoverable. However, if the recovery rate is less than 100%, the effect of allowing recourse depends on the cost of pursuing the deficiency judgment as well as the recovery rate—the effect of costs decreases as the recovery rate increases. As a result, the finding that recourse does not have a deterrent effect on default for low-value properties is consistent with costs being an important determinant of the effect of allowing recourse. Rate 5 Rate 4 Rate 2 LTV80 * Default Option LTV80 * Default Option Sq. Rate 1 LTV80 Default Option Sq. * Recourse Default Option Squared Default Option * Recourse Default Option 1.97 (0.19) –2.22 (0.21) –1.52 (0.35) 1.57 (0.43) 0.13 (0.02) 0.57 (0.19) –1.40 (0.30) –0.350 (0.025) –0.306 (0.033) 0.327 (0.011) 0.451 (0.015) All 1.65 (0.25) –1.59 (0.33) –0.10 (0.34) –0.26 (0.50) 0.05 (0.01) 0.71 (0.29) –1.67 (0.27) –0.121 (0.038) –0.166 (0.019) 0.259 (0.018) 0.423 (0.023) 2.14 (0.23) –2.36 (0.27) –0.82 (0.46) 1.00 (0.58) 0.06 (0.02) 0.59 (0.17) –1.29 (0.25) –0.122 (0.050) –0.157 (0.021) 0.296 (0.021) 0.487 (0.042) Fannie Mae (FNMA) All Appraisal >$200,000 1.99 (0.27) –2.38 (0.31) –0.16 (0.37) –0.33 (0.50) 0.07 (0.01) 0.85 (0.18) –1.83 (0.47) –0.165 (0.073) –0.144 (0.023) 0.244 (0.018) 0.481 (0.029) 2.16 (0.21) –2.56 (0.31) 0.07 (0.33) –0.85 (0.87) 0.10 (0.02) 0.79 (0.20) –1.59 (0.49) –0.237 (0.110) –0.153 (0.035) 0.285 (0.039) 0.560 (0.058) Freddie Mac (FHMLC) All Appraisal >$200,000 Downloaded from rfs.oxfordjournals.org by guest on September 7, 2011 Table 8 Recourse and the probability of default by investor type 2.10 (0.18) –2.56 (0.21) –2.08 (0.41) 2.29 (0.51) 0.15 (0.02) 0.27 (0.23) –0.97 (0.38) –0.448 (0.031) –0.414 (0.033) 0.285 (0.012) 0.353 (0.011) Private Securitized (continued) 1.90 (0.18) –1.71 (0.17) –1.15 (0.31) 1.07 (0.37) 0.17 (0.01) 0.39 (0.19) –1.19 (0.37) –0.270 (0.059) –0.308 (0.041) 0.397 (0.015) 0.632 (0.027) Private Portfolio Recourse and Residential Mortgage Default: Evidence from US States 3163 0.004 (0.023) 0.025 (0.025) –0.307 (0.014) 0.205 (0.048) –0.127 (0.073) 0.147 (0.014) 0.011 (0.001) 0.088 (0.007) 0.017 (0.013) 0.0190% 37,577,506 0.002 (0.016) 0.011 (0.025) –0.327 (0.011) 0.278 (0.046) –0.089 (0.078) 0.167 (0.017) 0.012 (0.001) 0.095 (0.011) 0.003 (0.022) 0.0113% 21,452,670 Fannie Mae (FNMA) All Appraisal >$200,000 0.030 (0.017) 0.004 (0.018) –0.232 (0.015) 0.130 (0.023) –0.047 (0.020) 0.280 (0.010) 0.009 (0.001) 0.075 (0.008) –0.163 (0.017) 0.165% 17,196,843 0.024 (0.015) 0.002 (0.020) –0.284 (0.018) 0.332 (0.059) –0.118 (0.064) 0.094 (0.042) 0.010 (0.001) 0.125 (0.012) 0.037 (0.035) 0.0096% 12,897,217 0.010 (0.018) 0.022 (0.020) –0.273 (0.016) 0.307 (0.053) –0.155 (0.060) 0.080 (0.028) 0.010 (0.001) 0.124 (0.009) 0.063 (0.013) 0.0142% 22,688,658 Private Securitized Freddie Mac (FHMLC) All Appraisal >$200,000 0.026 (0.019) 0.014 (0.023) –0.182 (0.028) 0.231 (0.030) –0.004 (0.022) 0.261 (0.021) 0.008 (0.001) 0.125 (0.007) –0.082 (0.014) 0.0585% 7,460,381 Private Portfolio Dependent variable in the probit is a binary variable that equals 1 if the loan defaults in that month, 0 otherwise. The benchmark specification is specification (3) from Table 3 (recourse in interactions). Default Option is the probability that the borrower has negative home equity. LTV80 = 1 if the loan has an LTV of exactly 80%, 0 otherwise. Rate variables control for the difference between the current mortgage rate and the contract rate. The results for Fico Score show the effect of a 100-point increase in the FICO score. Purpose Type Dummy = 1 if the loan is not a purchase mortgage, 0 otherwise. % Defaults is the % of observations that are defaults. Standard errors (clustered by state) are in parentheses. All regressions include a constant. Purpose Type Dummy % Defaults Number of obs. LTV Ratio at Origination Ln Loan Age ARM Dummy Lagged Unemp Rate Fico Score at Origination Interest Only Dummy Jumbo Dummy 0.023 (0.014) 0.007 (0.020) –0.264 (0.015) 0.187 (0.022) 0.027 (0.017) 0.310 (0.012) 0.010 (0.001) 0.077 (0.005) –0.099 (0.021) 0.050% 85,888,286 All Downloaded from rfs.oxfordjournals.org by guest on September 7, 2011 3164 Divorce Rate Table 8 Continued The Review of Financial Studies / v 24 n 9 2011 Recourse and Residential Mortgage Default: Evidence from US States of the probabilities for recourse and non-recourse states. At the mean value of the default option at the time of default and for securitized, privately held loans, borrowers in non-recourse states are 45% more likely to default than borrowers in recourse states. For privately held portfolio loans, borrowers in non-recourse states are 44% more likely to default.10 The estimation results in Table 8 indicate that recourse does not have a significant deterrent effect on default for loans held by FNMA or FHMLC. The coefficients on the interaction between the default option value and recourse for the FNMA and FHMLC samples are much smaller in magnitude than those for privately securitized loans, and they are statistically insignificant. This is true even when we consider only FNMA and FHMLC loans on properties appraised at $200,000 or more (in real 2005 dollars), the threshold above which we find that recourse matters. We conclude that recourse has a statistically significant deterrent effect on default only for privately held loans. 4. The Impact of Recourse on the Default Method We next turn to the question of how lender recourse affects the way in which default occurs. First, we examine whether recourse makes it more likely that a borrower will default through a deed in lieu or a short sale. To the extent that deeds in lieu and short sales represent more lender-friendly defaults than foreclosures, we expect defaults to occur more frequently by these methods in recourse states because lenders can use the threat of a deficiency judgment as a negotiating tool. In this respect, we envision a bilateral bargaining model 10 We also estimate our benchmark specification on FHA and VA loans. As these loans are explicitly non-recourse in all states, we should not find a significant negative coefficient on the interaction between recourse and the default option value. We find that the coefficient on the interaction between recourse and the default option value is positive, albeit small, for FHA and VA loans. 3165 Downloaded from rfs.oxfordjournals.org by guest on September 7, 2011 3.5 Recourse and expected LTVs Table 9 presents the results from the probit regression when the current expected LTV is used as a proxy for the default option, rather than the probability of negative equity as in our benchmark specification. The expected LTVs are computed using average home prices. The results are similar to those we obtain from our benchmark specification. In the full sample, the coefficient on the interaction between recourse and the expected LTV is statistically significant at the 10% level. The effect is highly significant for higher appraisal values. In Table 10, we use the results from Table 9 to compute the increase in the expected LTV in a recourse state relative to the expected LTV in a nonrecourse state, such that the default probability in both states is the same. In the full sample, the effect of recourse is to increase the expected LTV at which a borrower defaults by 8.6%. The equivalent expected LTV in a recourse state rises to 22% for mortgages on properties appraised at $300,000 to $500,000 and to 24% for mortgages on properties appraised at $500,000 to $750,000. 3166 Lagged Unemp Rate Fico Score at Origination Interest Only Dummy Jumbo Dummy Divorce Rate Rate 5 Rate 4 Rate 2 LTV80 * Expected LTV Rate 1 Expected LTV * Recourse LTV80 Expected LTV < $100,000 0.0170 (0.0017) 0.00067 (0.00050) –0.034 (0.046) 0.00095 (0.00063) –0.163 (0.099) –0.143 (0.031) 0.231 (0.014) 0.343 (0.021) 0.011 (0.016) 0.008 (0.015) –0.199 (0.018) –0.041 (0.034) – All 0.0141 (0.0009) 0.00112 (0.00067) 0.067 (0.069) 0.00062 (0.00071) –0.280 (0.022) –0.257 (0.026) 0.315 (0.009) 0.429 (0.014) 0.014 (0.014) –0.040 (0.029) –0.266 (0.015) –0.137 (0.022) 0.010 (0.015) 0.0135 (0.0019) –0.00019 (0.00043) –0.053 (0.038) 0.00172 (0.00054) –0.240 (0.048) –0.199 (0.023) 0.251 (0.012) 0.343 (0.018) –0.018 (0.016) –0.018 (0.018) –0.274 (0.013) 0.096 (0.021) – $100,000 to $200,000 0.0160 (0.0012) –0.00154 (0.00062) 0.019 (0.038) 0.00131 (0.00045) –0.261 (0.039) –0.265 (0.033) 0.298 (0.017) 0.423 (0.019) 0.012 (0.017) –0.066 (0.026) –0.293 (0.012) 0.145 (0.014) – $200,000 to $300,000 $500,000 to $750,000 0.0172 (0.0017) –0.00333 (0.00117) 0.413 (0.057) –0.00333 (0.00064) –0.426 (0.066) –0.407 (0.033) 0.354 (0.021) 0.555 (0.035) 0.076 (0.028) –0.143 (0.054) –0.300 (0.027) 0.190 (0.019) 0.047 (0.036) $300,000 to $500,000 0.0160 (0.0014) –0.00288 (0.00102) 0.217 (0.048) –0.00082 (0.00056) –0.264 (0.044) –0.264 (0.038) 0.354 (0.029) 0.478 (0.025) 0.047 (0.019) –0.110 (0.045) –0.297 (0.019) 0.154 (0.011) –0.033 (0.012) Downloaded from rfs.oxfordjournals.org by guest on September 7, 2011 Table 9 Recourse, expected LTVs, and the probability of default 0.0182 (0.0018) –0.00207 (0.00082) 0.497 (0.091) –0.00460 (0.00109) –0.223 (0.076) –0.284 (0.076) 0.473 (0.063) 0.635 (0.075) 0.035 (0.028) –0.129 (0.045) –0.275 (0.022) 0.170 (0.028) – $750,000 to $1,000,000 (continued) 0.0136 (0.0019) –0.00037 (0.00087) 0.247 (0.119) –0.00253 (0.00134) –0.107 (0.079) –0.215 (0.049) 0.435 (0.030) 0.642 (0.060) 0.018 (0.029) –0.069 (0.036) –0.281 (0.027) 0.137 (0.051) – > $1,000,000 The Review of Financial Studies / v 24 n 9 2011 < $100,000 −0.245 (0.014) −0.0083 (0.0015) 0.209 (0.016) −0.054 (0.016) −3.16 (0.15) 0.115% −55,102 12.52% 7,076,266 All −0.272 (0.008) −0.0011 (0.0012) 0.194 (0.012) −0.101 (0.024) −2.86 (0.16) 0.050% −307,783 17.36% 85,888,286 −0.299 (0.012) −0.0025 (0.0019) 0.202 (0.017) −0.076 (0.015) −2.71 (0.15) 0.048% −95,289 15.32% 26,973,983 $100,000 to $200,000 −0.314 (0.015) −0.0035 (0.0012) 0.221 (0.016) −0.104 (0.023) −2.60 (0.17) 0.039% −56,417 18.71% 20,004,481 $200,000 to $300,000 −0.293 (0.010) 0.0004 (0.0011) 0.226 (0.031) −0.165 (0.035) −2.70 (0.24) 0.046% −62,605 21.39% 19,840,252 $300,000 to $500,000 Downloaded from rfs.oxfordjournals.org by guest on September 7, 2011 −0.162 (0.024) 0.0000 (0.0014) 0.216 (0.038) −0.169 (0.028) −2.61 (0.26) 0.050% −26,170 21.32% 7,675,467 $500,000 to $750,000 −0.159 (0.030) −0.0036 (0.0029) 0.177 (0.034) −0.108 (0.022) −2.49 (0.22) 0.034% −5,786 19.40% 2,370,213 $750,000 to $1,000,000 −0.087 (0.045) 0.0003 (0.0019) 0.159 (0.025) −0.059 (0.027) −2.57 (0.25) 0.027% −4,085 15.31% 1,947,624 > $1,000,000 The dependent variable in the probit is a binary variable that equals 1 if the loan defaults in that month, 0 otherwise. The expected LTV is the current LTV (in %) computed using average state home prices. LTV80 takes a value of 1 if the loan has an LTV of exactly 80%, 0 otherwise. The rate variables control for the difference between the current mortgage rate and the contract rate. The results for Fico Score at Origination show the effect of a 100-point increase in the FICO score. Purpose Type Dummy takes a value of 1 if the loan is not a purchase mortgage, 0 otherwise. % Defaults is the percentage of observations that are defaults. Standard errors (clustered by state) are in parentheses. Robustness results are available from the authors. Coefficients in bold are significant at the 5% level. % Defaults Log ps. likelihood Pseudo R-sq Number of obs. Purpose Type Dummy Constant LTV Ratio at Origination Ln Loan Age ARM Dummy Table 9 Continued Recourse and Residential Mortgage Default: Evidence from US States 3167 The Review of Financial Studies / v 24 n 9 2011 Table 10 Increase in expected LTV for default in recourse states relative to non-recourse states All 8.6% < $100,000 $100,000 to $200,000 $200,000 to $300,000 $300,000 to $500,000 $500,000 to $750,000 $750,000 to $1,000,000 > $1,000,000 −3.8% 1.5% 10.7% 22.0% 24.0% 12.8% 2.8% This table uses the coefficients from Table 9 to compute the increase in the expected LTV in a recourse state relative to that in a non-recourse state. The interpretation is that borrowers in recourse states, on average, default at an expected LTV that is x% higher than the expected LTV at which a borrower would default at in a nonrecourse state. in which recourse affects the borrower’s and lender’s payoffs from defaulting.11 The results corroborate our hypothesis. We then look at whether recourse makes it more likely that borrowers will cure their defaults. We find that the cure rate is higher in recourse states. Finally, we find that borrowers that default in recourse states are more likely to attempt to resume payments during the delinquency period preceding the foreclosure. 11 The full model and its solution are available from the authors. 12 We calculate the partial effects at the mean of continuous variables and at the modes of dummy variables. 3168 Downloaded from rfs.oxfordjournals.org by guest on September 7, 2011 4.1 Recourse, deeds in lieu, and short sales We estimate a probit model to determine which factors influence whether borrowers are more likely to default by foreclosure than by a deed in lieu or a short sale. The sample is restricted to the observations for which the default variable takes a value of 1. The dependent variable takes a value of 1 if the default is by a foreclosure and 0 otherwise. Borrowers may be less likely to default by foreclosure in states with recourse. They may also be more likely to default by friendly foreclosure than by contested foreclosure in recourse states. However, we are unable to empirically distinguish between friendly foreclosures and contested foreclosures. To test the hypothesis that recourse influences how the borrower defaults, we include a recourse variable dummy as an explanatory variable for the probability of default by foreclosure. As controls, we include the borrower’s FICO score at origination and the LTV at origination to control for unobserved heterogeneity in either the borrower’s costs of decreased access to credit or the borrower’s search costs. Column 1 of Table 11 contains the results of the estimation, which indicate that recourse lowers the probability of default by foreclosure. The estimated coefficient is negative and statistically significant. In particular, the probability of default by foreclosure in recourse states is 10% lower than the corresponding probability in non-recourse states.12 The effect of recourse on the way a borrower defaults depends on how much recourse the lender has, as well as the LTV at the time of default. However, the relationships are nonlinear. For high LTVs and high recovery rates, the deterrent effect of a deficiency judgment is strong enough to deter default Recourse and Residential Mortgage Default: Evidence from US States Table 11 Recourse, deeds in lieu, and short sales (1) (2) (3) (4) (5) −0.527 (0.164) – −0.484 (0.155) −0.718 (0.117) −0.485 (0.458) −0.003 (0.034) 0.009 (0.003) 0.090 (0.287) – −0.008 (0.028) 0.003 (0.002) – – – −0.526 (0.137) −0.694 (0.139) −0.497 (0.457) 0.003 (0.035) 0.008 (0.003) 0.097 (0.288) – Appraisal Amount – – – −0.523 (0.138) −0.667 (0.142) −0.514 (0.461) 0.020 (0.042) 0.007 (0.003) 0.097 (0.289) −0.053 (0.052) – Appraisal Amount * Recourse Constant – – – – 0.85 (0.30) 86.2% −16,777 3.58% 43,353 0.79 (0.38) 86.2% −16,761 3.68% 43,353 0.77 (0.38) 86.4% −16,556 3.78% 43,252 Default Option % Foreclosures Log ps. likelihood Pseudo R-squared Number of obs. – 1.26 (0.30) 86.2% −16913.797 2.80% 43,353 0.030 (0.019) −0.017 (0.040) 0.75 (0.36) 86.2% −16,756 3.71% 43,353 The dependent variable in the probit is a binary variable that takes a value of 1 if default is by foreclosure, 0 otherwise. Default Option is the probability that the borrower has negative home equity. Recourse is a dummy variable that takes a value of 1 if the property is in a recourse state, 0 otherwise. The coefficients and standard errors for Fico Score at Origination show the effect of a 100-point increase in the FICO score. Foreclosure Timing dummy is a dummy variable that takes a value of 1 if the uncontested foreclosure time is less than six months. Investor Type 1 is a dummy variable that takes a value of 1 if the lender type is not “Private Portfolio” or “Private Securitized,” 0 otherwise. Appraisal amount is the appraisal amount of the property at origination; coefficients and standard errors shown are for the effect of a $100,000 increase. The number of observations in column 4 differs from the number of observations in other columns because observations with investor type “Unknown” are excluded for this specification. In all specifications, standard errors are clustered by state. Coefficients in bold are significant at the 5% level. altogether. For high LTVs and more moderate recovery rates, it is sometimes worthwhile for the lender to pursue a deficiency judgment and for the borrower to default. For moderate LTVs and low-to-moderate recovery rates, recourse changes how the borrower defaults, rather than whether the borrower defaults. To test whether recourse has a stronger effect for higher values of the default option, we add the default option value and the default option value interacted with the recourse dummy, in addition to the recourse variable, as explanatory variables for the probability of default by foreclosure. If recourse has a stronger negative effect at higher values of the default option, we would expect a negative coefficient on the interaction term between recourse and the default option value. As shown in column 2 of Table 11, the negative effect of recourse on the probability of default by foreclosure is stronger for higher values of the default option. However, the effect is not statistically significant. 3169 Downloaded from rfs.oxfordjournals.org by guest on September 7, 2011 Default Option * Recourse Fico Score at Origination LTV Ratio at Origination Foreclosure Timing Dummy Investor Type 1 −0.554 (0.158) −0.679 0.133 −0.523 (0.525) 0.005 (0.032) 0.008 (0.003) – Recourse The Review of Financial Studies / v 24 n 9 2011 4.2 Cure rates To examine whether borrowers in recourse states are more likely to remedy their delinquencies, we use the methodology developed by Adelino, Gerardi, and Willen (2009). A cured loan is one that is prepaid, current, or 30 days delinquent 12 months after the initial 60-day delinquency. As in Adelino, Gerardi, and Willen (2009), we restrict our analysis to the first 60-day delinquency and we do not look at cures on any subsequent 60-day delinquencies. In the full sample, 42% of loans in non-recourse states are cured, while 57% of loans in recourse states are cured. We also examine cure rates in different product categories. For interest-only loans, the cure rates in non-recourse and recourse states are 25% and 35%, respectively. For adjustable-rate mortgages, the cure rates in non-recourse and recourse states are 36% and 49%, respectively. The cure rates for fixed-rate mortgages are 50% in non-recourse states and 62% in recourse states. We also look at whether other loan and borrower 3170 Downloaded from rfs.oxfordjournals.org by guest on September 7, 2011 Theory does not provide a clear prediction regarding the effect of the time it takes to foreclose on the share of lender-friendly defaults. On the one hand, a longer foreclosure process makes it more likely that the lender will prefer a lender-friendly default to a foreclosure and will forgo a deficiency judgment in favor of a deed in lieu or a short sale. On the other hand, the borrower will prefer foreclosure when he can delay the search and credit costs, and receive a longer period of free rent as a result of a longer foreclosure process. A priori, the effect of foreclosure timing on the process is unclear. To empirically examine the effect, we include a dummy variable that takes a value of 1 if the uncontested foreclosure time is less than six months and 0 otherwise. As the results in column 3 of Table 11 indicate, foreclosure timing does not have a significant effect on the probability of default by foreclosure: The partial effect evaluated at the means implies an increase in probability of 2% and is far from statistically significant. The results are similar when we include foreclosure timing as a continuous variable rather than as a dummy variable. Finally, we examine whether lender type and appraisal amount affect the probability of defaulting by litigious foreclosure. To examine the effect of lender type, we include a dummy variable that takes a value of 1 if the lender is a GSE and 0 otherwise, i.e., when the loan is privately securitized or held in a private lender’s portfolio. As is evident in column 4 of Table 11, mortgages held by a GSE are no more likely to default through foreclosure than mortgages held by private lenders. To examine the effect of the appraisal amount of the property on the probability of default by foreclosure, we include the appraisal amount and the appraisal amount interacted with the recourse dummy as explanatory variables. We present the estimation results in column 5 of Table 11. The coefficient on the appraisal amount is positive and marginally significant. The coefficient on the interaction term with recourse is negative but statistically insignificant. Recourse and Residential Mortgage Default: Evidence from US States 4.3 Recourse and direct default We now look at whether borrowers who default in non-recourse states are more likely than borrowers in recourse states to default directly in the sense of abruptly stopping payments without resuming them at a later point. Foote, Gerardi, Goette, and Willen (2009) call such defaults “direct defaults” and suggest that these payment patterns are more prevalent among borrowers who default strategically. The intuition is that, in the absence of a sudden change in home price expectations, a borrower who decides to strategically default will simply cease making payments on the mortgage. If, instead, a borrower is defaulting because of a liquidity constraint, that borrower may resume payments if the constraint is relaxed following the initial delinquency. For example, a direct defaulter has a monthly payment pattern, such as CCC3699999, where C denotes current in that month, 3 denotes 30 days delinquent in that month, 6 denotes 60 days delinquent in that month, and 9 denotes 90+ days delinquent in that month. Conversely, a defaulter who has a payment pattern such as (a) CC333666999 or (b) CC36369F, where F denotes in foreclosure in that month, is much less likely to be a strategic defaulter. In case (a), which is a non-strategic case, the fact that the borrower is three or six months delinquent in a row indicates that, despite having fallen behind on the mortgage, the borrower keeps sending monthly payments to the lender. This borrower loses the home because he cannot keep up with payments. In case (b), the borrower falls two months behind on his mortgage and then makes two payments in one month, such that the lender codes him as 30-day delinquent once again. To identify direct defaults, we focus on our sample of defaulted loans, where default is defined as a default that ends in the borrower losing the home. 3171 Downloaded from rfs.oxfordjournals.org by guest on September 7, 2011 characteristics explain the difference in cure rates between non-recourse and recourse states. The difference between cure rates remains after we control for such characteristics, although the effect is not always statistically significant. The fact that borrowers are more likely to cure their delinquencies in recourse states may indicate that some borrowers are not aware that the mortgage is recourse until after they become delinquent. After the initial delinquency, they may learn that their loan is recourse from the servicer of the loan or from a foreclosure attorney. Indeed, in delinquency notifications, servicers generally inform the borrower that a mortgage is recourse if that is the case. After learning that the cost of default is higher than first anticipated, the borrower may make a greater effort to become current or decide against a strategic default. However, it is also possible that the higher cure rate for recourse loans reflects the fact that a greater share of defaults in recourse states is driven by liquidity constraints—if the borrower’s liquidity constraints are relaxed after the initial delinquency, the cure rate will be higher in recourse states. However, we cannot disentangle these two possibilities. Both channels are probably important in explaining the higher cure rate in recourse states. The Review of Financial Studies / v 24 n 9 2011 13 Foote, Gerardi, Goette, and Willen (2009) do not define exactly what they mean by seriously delinquent for this criterion, but we assume they exclude loans that were at least 90 days delinquent at some point in the past. 3172 Downloaded from rfs.oxfordjournals.org by guest on September 7, 2011 We first identify the month of a critical event. For loans that go into 90-day delinquent status prior to the borrower losing the home, the critical event is the first time the borrower goes 90-day delinquent with no subsequent resumption of payments prior to the borrower losing the home. Occasionally, we see a loan enter REO or F status without a prior 90-day delinquent status. In such cases, the borrower most likely loses the home through either deeds in lieu or friendly foreclosures. For such loans, we define the critical event as the time at which the borrower loses the home. We then define a direct default as a loan for which the payment history satisfies the following criteria: 1) in the six months prior to the critical event, the borrower records only one month of 30-day delinquency and only one month of 60-day delinquency; and 2) at no time in the loan’s history prior to the six-month stretch before the critical event was the borrower 60 or more days behind on his payments. Criterion 1 is identical to that used by Foote, Gerardi, Goette, and Willen (2009). Criterion 2 is a somewhat more stringent criterion than Foote et al.’s, as it excludes loans that were 60 days delinquent at some point in the past rather than only those loans that were “seriously delinquent” in the past.13 In our sample of defaulted loans, we find that 44% of defaults are direct defaults. In non-recourse states, 51% of defaults are direct defaults, while only 39% of defaults in recourse states are direct defaults. Unconditionally, the share of defaults that are direct is 31% higher (12 percentage points) in non-recourse states than in recourse states. In Table 12, we use a probit model to explore whether the higher share of direct defaults in non-recourse states is due to differences in loan characteristics, the default option value, the year in which default occurs, or differences in state foreclosure timelines. We include a dummy variable that takes a value of 1 if the critical event occurs in 2005 or later. The dependent variable takes a value of 1 if the default is direct and 0 otherwise. We find that the coefficient on recourse is negative and statistically significant. When controlling for the above characteristics, measured at the sample means, recourse decreases the likelihood that the default is direct by 6% in the full sample. The effect of recourse is statistically significant only for loans with appraisal amounts at origination of $300,000 to $750,000. Unlike our finding for the effect of recourse on default, the effect is not statistically significant for loans on properties appraised at $750,000 to $1 million at origination, although there are only 618 observations in this sample. Borrowers with higher FICO scores who default are more likely to default directly. This effect is highly statistically significant for the full sample and for every appraisal category. In this regard, a higher FICO score is the most robust predictor of direct default. This is consistent with an analysis by −0.030 (0.090) 0.252 (0.207) 0.434 (0.031) 0.00139 (0.00204) −0.0545 (0.0563) − −0.145 (0.049) 0.680 (0.065) 0.577 (0.026) 0.00111 (0.00095) 0.0924 (0.0477) 0.0134 (0.0058) 0.278 (0.044) −4.32 (0.18) −20,346 7.75% 32,172 0.181 (0.051) −3.38 (0.30) −3,415 3.93% 5,559 < $100,000 All 0.283 (0.065) −4.03 (0.18) −5,805 5.54% 9,347 −0.030 (0.061) 0.573 (0.192) 0.530 (0.021) 0.00008 (0.00158) 0.0213 (0.0513) − $100,000 to $200,000 0.139 (0.089) −4.74 (0.29) −3,812 7.45% 5,988 −0.104 (0.075) 0.726 (0.186) 0.641 (0.027) 0.00196 (0.00269) 0.1938 (0.0680) − $200,000 to $300,000 0.431 (0.109) −5.01 (0.21) −4,622 7.55% 7,227 −0.122 (0.059) 0.664 (0.101) 0.654 (0.036) 0.00164 (0.00262) 0.1927 (0.0707) − $300,000 to $500,000 Downloaded from rfs.oxfordjournals.org by guest on September 7, 2011 0.192 (0.192) −4.64 (0.27) −1,949 6.17% 3,042 −0.131 (0.052) 0.717 (0.074) 0.606 (0.022) 0.00367 (0.00317) 0.2246 (0.1108) − $500,000 to $750,000 0.442 (0.407) −4.49 (0.68) −409 4.08% 618 −0.063 (0.113) 0.218 (0.100) 0.523 (0.056) 0.00886 (0.00556) −0.1078 (0.2344) − $750,000 to $1,000,000 0.287 (0.306) −7.24 (0.90) −237 10.69% 391 0.140 (0.162) 0.273 (0.198) 0.832 (0.092) 0.01433 (0.00487) 0.4647 (0.2019) − > $1,000,000 The dependent variable in the probit is a binary variable that takes a value of 1 if the default is direct , 0 otherwise. A direct default refers to a default in which there is no evidence that the borrower tried to resume payments prior to losing the home; see text for full details. Default Option is the probability that the borrower has negative equity. Recourse is a dummy variable that takes a value of 1 if the property is in a recourse state, 0 otherwise. The coefficients and standard errors for Fico Score at Origination show the effect of a 100-point increase in the FICO score. Foreclosure Timing is a dummy variable that takes a value of 1 if the uncontested foreclosure time is less than six months. Appraisal amount is the appraisal amount of the property at origination; coefficients and standard errors shown are for the effect of a $100,000 increase. Standard errors are clustered by state. Coefficients in bold are significant at the 5% level. Robustness results are available from the authors. Log ps. likelihood Pseudo R-sq Number of obs. Constant Post 2004 Dummy Appraisal Amount Foreclosure Timing Dummy LTV Ratio at Origination Fico Score at Origination Default Option Recourse Table 12 Direct defaults and recourse Recourse and Residential Mortgage Default: Evidence from US States 3173 The Review of Financial Studies / v 24 n 9 2011 Experian-Oliver Wyman (2009), which finds that prime borrowers are more likely to strategically default than non-prime borrowers. The effect is quantitatively important: a 100-basis-point increase in the borrower’s FICO score at origination makes it 20% more likely that any default that does occur will be direct. We also find that direct defaults are more frequent from 2005 onward and that borrowers who have more expensive properties, measured in terms of the real appraisal amount at origination, are more likely to default strategically. 5. Is Recourse Priced? 3174 Downloaded from rfs.oxfordjournals.org by guest on September 7, 2011 To the extent that borrowers in recourse states are less likely to default in response in negative equity and are more likely to default in a lender-friendly way if they do default, lenders are likely to face smaller losses from default in recourse states. Thus, one might expect to see lower interest rates in recourse states. To explore this issue, we regress the contract rate at origination for all fixed-rate mortgages in our sample on the recourse dummy variable, our main controls, and year of origination dummies. Column 1 of Table 13 presents the results of this regression for the combined sample of purchase and nonpurchase mortgages. The coefficient on recourse is positive and statistically significant, indicating that borrowers in recourse states actually face higher interest rates. We find the same result when we estimate the effect of recourse on loan pricing in the sample of purchase mortgages (column 2) and in the sample of non-purchase mortgages (column 3). As our earlier results indicate that recourse only has an effect on mortgages on more expensive properties, we estimate the relationship between recourse and interest rates on mortgages on properties with different appraisal values at origination. Table 14 presents these results. Recourse continues to have a positive and statistically significant effect on the mortgage rate for properties appraised at $750,000 to $1,000,000. We found that recourse had a strong deterrent effect for mortgages on properties in this value range. In no appraisal category do we see evidence that borrowers in recourse states enjoy lower interest rates. We also find that rates on privately held mortgages are no lower in recourse states than rates on privately held mortgages in non-recourse states. What might explain this perplexing finding? First, recall that we find no evidence that recourse states have lower default rates when we include recourse only in levels in our regression. Thus, it is perhaps not entirely surprising that our rate regression, which also includes recourse only in levels, does not show that recourse lowers the default rate. Lenders may also have attached a very low probability to the state of the world in which home prices fall enough that middle- and upper-income homeowners, the types of households for which recourse deters default, default on their mortgages. However, the fact that the coefficient is positive and statistically significant suggests there may be important differences across states that are unrelated to recourse. For example, there may be differences in the amount of lender competition across states. Furthermore, Recourse and Residential Mortgage Default: Evidence from US States Table 13 Recourse and mortgage interest rates (fixed-rate mortgages only) Purchase only (2) Refi only (3) 6.02 (1.42) 1.29 (0.17) 0.262 (0.005) 20.5 (1.0) 0.76 (0.72) 3.46 (0.55) 0.024 (0.183) 0.265 (0.042) −0.71 (1.46) −0.378 (0.026) 13.1 (3.8) 49.8 (5.7) 0.28 (1.36) 56.41% 2,254,676 4.18 (1.90) 1.06 (0.15) 0.237 (0.004) 23.2 (1.0) 2.46 (0.91) 1.89 (0.62) 0.622 (0.195) 0.137 (0.032) 0.72 (0.83) −0.306 (0.015) 24.3 (5.3) 41.5 (4.8) 6.66 (1.66) 1.46 (0.21) 0.266 (0.005) 18.7 (1.3) (0.53) (1.03) 4.54 (0.69) (0.391) (0.331) 0.355 (0.062) −3.47 (2.01) −0.403 (0.034) 3.6 (2.3) 55.2 (6.9) 57.85% 781,814 60.69% 1,472,862 Recourse Loan Amt. at Orig Loan Term (m) Aver. State Int. Rate Divorce Rate Lagged Unemp Rate Foreclosure Timing (m) LTV Ratio at Origination LTV80 Fico Score at Origination Interest Only Dummy Jumbo Dummy Purpose Type Dummy R-squared Number of obs. The dependent variable is the current mortgage interest rate. LTV80 takes a value of 1 if the loan has an LTV of exactly 80%, 0 otherwise. Purpose Type Dummy takes a value of 1 if the loan is not a purchase mortgage, 0 otherwise. Standard errors clustered by state are in parentheses. The coefficients and standard errors for all variables except the loan amount are multiplied by 100; the coefficient and the standard error on the loan amount variable are multiplied by 1,000,000. Coefficients in bold are significant at the 5% level. Additional controls include a full set of year of origination dummies. our data do not include any information on origination points. Lenders may be pricing recourse using fees at origination rather than the contract rate. There may also be differences in lenders’ expectations of future home price appreciation.14 All of these explanations are plausible, and additional work is needed to disentangle them. 6. Conclusions and Discussion We empirically investigate the effect of recourse on default. We find that, in a sample of loans originated between August 1997 and December 2008, at the 14 We explored some specifications for the interest rate in which we had measures of future expected home price appreciation. In no specification did we find that recourse lowered the mortgage interest rate. However, our measures of future expected home price appreciation were based on lagged home prices and were probably not close approximations of lenders’ actual expectations. 3175 Downloaded from rfs.oxfordjournals.org by guest on September 7, 2011 Purchase and Refi (1) 48.17% 198,693 R-squared Number of obs. 55.75% 748,421 1.98 (1.54) 2.81 (3.00) −3.36 (0.14) 0.284 (0.002) 21.0 (0.9) 0.07 (0.74) 2.51 (0.61) 0.134 (0.147) 0.445 (0.047) −1.61 (1.50) −0.378 (0.013) 21.7 (2.4) 60.25% 552,054 2.26 (1.97) −1.18 (0.07) 0.284 (0.002) 23.9 (1.0) 0.71 (1.03) 1.58 (0.70) 0.303 (0.287) 0.123 (0.029) 1.17 (0.78) −0.302 (0.013) 22.9 (1.7) 18.7 (27.3) (0.91) (0.90) 61.35% 514,411 2.31 (1.68) −0.66 (0.05) 0.284 (0.002) 26.8 (1.3) 0.62 (0.94) 2.19 (0.79) 0.274 (0.317) 0.038 (0.021) 4.77 (1.04) −0.256 (0.010) 18.1 (2.6) 29.7 (1.0) (1.31) (0.65) $300,000 to $500,000 Purchase and Refi by Appraisal 5.33 (1.65) 0.11 (0.02) 0.278 (0.004) 24.1 (1.2) 3.95 (1.00) 2.25 (0.54) 0.580 (0.150) 0.085 (0.025) 7.61 (1.88) −0.175 (0.017) −4.0 (1.8) 11.4 (1.1) (0.42) (0.70) 3.19 (1.38) −0.12 0.05 0.276 (0.002) 23.1 (1.5) 2.34 (0.87) 2.28 (0.53) 0.515 (0.232) (0.029) (0.046) 10.51 (0.98) −0.193 (0.007) 4.6 (3.5) 22.9 (1.5) (1.80) (0.78) 64.33% 43,962 2.26 (1.33) −0.56 0.04 0.277 (0.002) 25.7 (1.1) 1.69 (0.87) 2.39 (0.55) 0.310 (0.228) 0.020 (0.018) 8.94 (1.27) −0.215 (0.005) 3.5 (3.9) 31.4 (0.7) 1.09 (0.53) 62.35% 168,320 63.41% 28,815 > $1,000,000 $750,000 to $1,000,000 $500,000 to $750,000 The dependent variable is the current mortgage interest rate. LTV80 takes a value of 1 if the loan has an LTV of exactly 80%, 0 otherwise. Purpose Type Dummy takes a value of 1 if the loan is not a purchase mortgage, 0 otherwise. Standard errors clustered by state are in parentheses. The coefficients and standard errors for all variables except the loan amount are multiplied by 100; the coefficient and the standard error on the loan amount variable are multiplied by 1,000,000. Coefficients in bold are significant at the 5% level. Additional controls include a full set of year of origination dummies. 19.41 (2.88) 9.05 (3.00) −14.40 (1.42) 0.289 (0.008) 13.9 (0.9) (2.45) (2.15) 6.43 (1.07) −1.509 (0.522) 1.393 (0.105) −6.62 (2.99) −0.628 (0.040) 21.8 (4.1) Purpose Type Dummy Jumbo Dummy Interest Only Dummy Fico Score at Origination LTV80 LTV Ratio at Origination Foreclosure Timing (m) Lagged Unemp Rate Divorce Rate Aver. State Int. Rate Loan Term (m) Loan Amt. at Orig Recourse < $100,000 $200,000 to $300,000 Downloaded from rfs.oxfordjournals.org by guest on September 7, 2011 3176 $100,000 to $200,000 Table 14 Recourse and mortgage interest rates (fixed-rate mortgages only) by appraisal The Review of Financial Studies / v 24 n 9 2011 Recourse and Residential Mortgage Default: Evidence from US States Appendix A: Foreclosure Laws by State Alabama: Lenders may foreclose through either a judicial or a non-judicial procedure. State law permits deficiency judgments without significant restrictions. We classify Alabama as a RECOURSE state. The borrower retains a right of redemption for one year after foreclosure. The relevant statutes are found in Section 35-10 of the Alabama Code. Alaska: Lenders may foreclose through either a judicial or a non-judicial procedure. The usual financing instrument is a deed of trust, and non-judicial foreclosure is the usual foreclosure process. State law permits deficiency judgments only if the lender pursues judicial foreclosure under the promissory note; no separate “deficiency judgment” is entered. The property sold through a judicial sale is subject to a right of redemption, and the redemption period is 12 months. As 3177 Downloaded from rfs.oxfordjournals.org by guest on September 7, 2011 mean value of the default option at the time of default, the probability of default is 32% higher in non-recourse states than in recourse states. The deterrent effect on default is significant only for borrowers with appraised property values of $200,000 or more at origination. At the mean value of the default option at the time of default and for homes appraised at $300,000 to $500,000, borrowers in non-recourse states are 81% more likely to default than borrowers in recourse states. For homes appraised at $500,000 to $750,000, borrowers in non-recourse states are more than twice as likely to default as borrowers in recourse states. For homes appraised at $750,000 to $1 million, borrowers in non-recourse states are 60% more likely to default. We also find that recourse deters default on loans held privately, but we cannot reject the hypothesis that recourse has no effect on loans held by government-sponsored enterprises. Finally, we find that allowing lenders recourse increases the likelihood that default occurs through a more lender-friendly method, such as a deed in lieu of foreclosure. Our findings shed light on the ongoing discussions about the existence of strategic default. The finding that recourse deters default indicates that at least some residential mortgage defaults are strategic rather than involuntary (i.e., the borrower has no choice but to default because of liquidity constraints). Our results indicate that some borrowers choose not to default when the lender has recourse, which indicates that they are capable of continuing to make payments on their mortgages. Our results regarding the differential effect of recourse by the mortgaged property’s appraisal amount indicate that at least some defaults on high and moderately priced homes are strategic. We cannot eliminate the possibility that some of the defaults on low-priced homes are strategic—the appraisal amount serves as a proxy for both the lender’s amount of recourse and the borrower’s financial means in general. Recourse may not significantly affect default on low-priced homes for one of two reasons. The first possibility is that most households with low-priced homes are liquidity constrained and therefore default because of their inability to carry payments. Alternatively, for low-priced properties, the lender’s recovery on a deficiency judgment may be low in practice because of a low recovery rate and a relatively higher fixed cost. The Review of Financial Studies / v 24 n 9 2011 3178 Downloaded from rfs.oxfordjournals.org by guest on September 7, 2011 judicial foreclosure is substantially more time consuming and cumbersome, we classify Alaska as a NON-RECOURSE state. The relevant statutes are found in Title 34, Chapter 20, Section 100 of the Alaska Statutes. Arizona: Lenders may foreclose through either a judicial or a non-judicial procedure. The usual financing instrument is a deed of trust, and non-judicial foreclosure is the usual foreclosure process. Deficiency judgments are not permitted if the property is residential, if it is located on 2.5 acres or less, and if its intended use is as a one- or two-family dwelling. We classify Arizona as a NON-RECOURSE state. The relevant statute is Article 33 of the Arizona State Code. Arkansas: Lenders may foreclose through either a judicial or a non-judicial procedure. Lenders usually foreclose on a deed of trust through a non-judicial procedure. State law permits deficiency judgments with the restriction that borrowers must receive credit for the greater of the foreclosure sale price or the fair market value of the property. We classify Arkansas as a RECOURSE state. The relevant statutes are found in Sections 18-50-212 and 18-50-216 of the Arkansas Code. California: Lenders may foreclose through either a judicial or a non-judicial procedure. Nonjudicial foreclosure is the usual foreclosure process. The borrower has five days to reinstate in a non-judicial foreclosure process. State law prohibits deficiency judgments on purchase mortgages. On other residential mortgages, state law permits deficiency judgments only if the lender pursues the more expensive and time-consuming judicial foreclosure process rather than the non-judicial foreclosure process. The lender may only file for a payment of the difference between the debt owed and the fair market value of the property. A deficiency suit also gives the borrower a right to redemption. We classify California as a NON-RECOURSE state. The relevant statutes are found in Sections 2920-2944.5 of the California Code. Colorado: Lenders may foreclose through either a judicial or a non-judicial procedure. Nonjudicial foreclosure is the norm. State law permits deficiency judgments. However, judges require lenders to bid fair market value on the property in the event that the total debt owed exceeds the property value less reasonable expenses. If the borrower can show that lenders bid less than fair market value, the borrower can avoid a deficiency judgment. After the sale, there is a redemption period of 75 days. There are no unreasonably burdensome statutory limitations on either filing or collecting on a deficiency or collection. We classify Colorado as a RECOURSE state. The relevant statutes are Title 38, Articles 37–39, of the Colorado Revised Statutes. Connecticut: Lenders may foreclose using one of two judicial procedures: a strict foreclosure and a decree of sale foreclosure. State law permits deficiency judgments under both procedures. However, if the lender pursues decree of sale foreclosure, the lender must first credit the borrower with one-half the difference between the debt and the appraised value if the property is sold pursuant to a court order and the property sells for less than the appraised value. In a strict foreclosure, the judge determines the fair market value of the property for which the borrower receives credit; a motion for deficiency judgment must be filed within 29 days of title vesting. There is no statutory deadline to file the motion for deficiency judgment after foreclosure-by-sale. We classify Connecticut as a RECOURSE state. The relevant statutes are Sections 49-14 and 49-28 of the General Statutes of Connecticut. Delaware: Lenders may foreclose only through a judicial procedure. State law permits deficiency judgments without significant restrictions. We classify Delaware as a RECOURSE state. The relevant statute is Title 10, Chapter 49:XI, of the Delaware Code. District of Columbia: Lenders may only foreclose through a non-judicial procedure. At any time within thirty days after the time limit for redemption has expired, any party to a mortgage foreclosure may file a motion seeking a deficiency judgment. We classify the District of Columbia as a RECOURSE district. The relevant statute is Title 42, Chapter 8, of the District of Columbia Code. Florida: Lenders may foreclose only through judicial foreclosure. State law permits deficiency judgments subject to the borrower receiving credit for the greater of the fair market value of the property or the foreclosure sale price. A deficiency judgment can be pursued against the original Recourse and Residential Mortgage Default: Evidence from US States 3179 Downloaded from rfs.oxfordjournals.org by guest on September 7, 2011 makers of a note even if they were not a party to the foreclosure action. However, Florida has an extremely generous homestead exemption, such that if the property is an investment property rather than a primary residence, the borrower can partially shield his or her assets from collection on the deficiency. We classify Florida as a RECOURSE state. The relevant statutes are found in Title 40, Chapter 702, of the Florida Statutes. Georgia: Lenders may foreclose through either a judicial or a non-judicial procedure. Nonjudicial foreclosure is the usual process. A prerequisite to a deficiency judgment is that the court has confirmed and approved the sale, which in turn requires that the sale price is equal to at least the fair market value of the property. The lender must receive such confirmation and approval within 30 days of the foreclosure sale. There is no right of redemption. We classify Georgia as a RECOURSE state. The relevant statutes are found in Title 44, Chapter 14, of the Official Code of Georgia. Hawaii: Lenders may foreclose through either a judicial or a non-judicial procedure. A judicial foreclosure takes 320 days, while a non-judicial foreclosure takes 195 days if it is uncontested. State law permits deficiency judgments if the lender pursues judicial foreclosure. The deficiency judgment process, if not contested, is fairly inexpensive. We classify Hawaii as a RECOURSE state. The relevant statutes are Chapters 667-5 and 667-38 of the Hawaii Revised Statutes. Idaho: Lenders may foreclose through either a judicial or a non-judicial procedure, although judicial foreclosure is exceptionally rare. State law permits a deficiency judgment provided the lender applies for one within 90 days of the foreclosure sale. The deficiency is limited to the difference between the balance owed and the fair market value of the property. The deficiency judgment process is onerous in practice, as the lender must prove fair market value, which the borrower can contest. We classify Idaho as a RECOURSE state. The relevant statutes are found in Title 45, Chapter 15, Section 45.12, of the Idaho Statutes. Illinois: Lenders may foreclose only through judicial foreclosure. State law permits deficiency judgments provided the borrower is personally served with the deficiency suit. Furthermore, a judge must confirm the sale and, according to Chapter 735, Article XV, Section 15-1508, ajudge may opt not to confirm the sale on the grounds that “justice was not otherwise done.” In practice, this means that the granting of a deficiency judgment is at the discretion of the judge and judges rarely grant deficiency judgments on residential property. We classify Illinois as a RECOURSE state because the possibility of personal recourse may be sufficient to deter some strategic defaulters, even if deficiency judgments are granted only rarely. The relevant statutes are found in Chapter 735, Article XV, of the Illinois Compiled Statutes. Indiana: Lenders may foreclose only through judicial foreclosure, which optimally takes 266 days if uncontested. State law permits deficiency judgments on residential properties without significant restrictions. The borrower must be served in person, which is not a significant restriction in practice. We classify Indiana as a RECOURSE state. The relevant statutes are found in Article 29, Chapter 7, of the Indiana State Code. Iowa: Lenders may foreclose only through judicial foreclosure. State law permits deficiency judgments on nonagricultural residential properties. However, seeking a deficiency judgment significantly delays the foreclosure process. Furthermore, there is a two-year statute of limitations on collecting on the deficiency judgment and generous limits on the garnishment of wages. The law makes it much faster to foreclosure on property if the lender waives the right to a deficiency judgment. As deficiencies are hard to collect in Iowa, lenders may even compensate a borrower who agrees to vacate the property quickly by paying the first month’s rent for new housing. We classify Iowa as a NON-RECOURSE state. The relevant statute is Chapter 654.6 of the Iowa Code. Kansas: Lenders may foreclose only through judicial foreclosure. After a foreclosure sale, a deficiency judgment is automatically entered if the sale proceeds less expenses are not sufficient to cover the debt owed. The borrower may contest the deficiency if the foreclosure sale price was less than the fair market value of the property. Kansas is unusual in that redemption rights can be sold to third parties. Therefore, if the lender bids substantially less for the property than its fair market value, the holder of the redemption rights may obtain the property at significantly less than The Review of Financial Studies / v 24 n 9 2011 3180 Downloaded from rfs.oxfordjournals.org by guest on September 7, 2011 market value. Furthermore, second lien holders lose the right to a deficiency if they do not ask for a foreclosure themselves. We classify Kansas as a RECOURSE state. The relevant statute is Chapter 60, 2417, of the Kansas Statutes. Kentucky: Lenders may foreclose only through judicial foreclosure. After a foreclosure sale, a deficiency judgment is automatically entered if the sale proceeds less expenses are not sufficient to cover the debt owed. There are no significant restrictions. We classify Kentucky as a RECOURSE state. The relevant statutes are found in Chapter 426 of the Kentucky Revised Statutes. Louisiana: Lenders may foreclose only through judicial foreclosure. State law permits deficiency judgments on residential properties without significant restrictions. We classify Louisiana as a RECOURSE state. The relevant statutes are in Title 10:9-629 of the Louisiana Code. Maine: Lenders may foreclose only through judicial foreclosure. State law permits deficiency judgments on residential properties provided the lender credits the borrower’s account for the fair market value of the property. We classify Maine as a RECOURSE state. The relevant statutes are found in Title 14, Part 4, Chapter 403, of the Revised Maine Statutes. Maryland: Lenders may foreclose through either a judicial or a non-judicial procedure. State law permits deficiency judgments on residential properties without significant restrictions. We classify Maryland as a RECOURSE state. The relevant statutes are found in Title 14, Chapter 200, of the Maryland Rules. Massachusetts: Lenders may foreclose through either a judicial or a non-judicial procedure. State law permits a deficiency judgment if the lender gives the borrower written notice prior to the foreclosure sale that the lender intends to pursue a deficiency. We classify Massachusetts as a RECOURSE state. The relevant statutes are found in Chapter 244 of the General Laws of Massachusetts. Michigan: Lenders may foreclose through either a judicial or a non-judicial procedure. There is typically a six-month redemption period after the completion of a non-judicial foreclosure. State law permits a deficiency judgment without significant restrictions in the case of judicial foreclosure. In the case of non-judicial foreclosure, the borrower can contest the deficiency if the property sold for substantially less than the fair market value. We classify Michigan as a RECOURSE state. The relevant statutes are found in the Michigan Compiled Laws, Chapter 451, and in the EPIC Act 236, Sections 600 and 700. Minnesota: Lenders may foreclose through either a judicial or a non-judicial procedure, although in the vast majority of cases, lenders foreclose through a non-judicial process. There are substantial redemption rights in Minnesota. In particular, the borrower is entitled to a six- or twelve-month redemption period after the foreclosure sale. The borrower is entitled to possession of the property, and the lender has limited rights to enter the property. The redemption period can be shortened to six months if certain conditions are met. A separate court procedure is required to shorten the redemption period to five weeks if the residential property is deemed “abandoned” and if it is composed of less than five units and is on less than ten acres. Thus, including the redemption period, the optimum timeframe for non-judicial foreclosure is 270–280 days. In the event the lender forecloses by advertisement, state law prohibits deficiency judgments. In judicial foreclosure, the lender may obtain a deficiency judgment subject to the borrower receiving credit for the fair market value of the property, which is determined by a jury. As judicial foreclosure is substantially more onerous than the non-judicial procedure, lenders pursue non-judicial foreclosure in the vast majority of cases. We classify Minnesota as a NON-RECOURSE state. The relevant statutes are in found in Chapters 580 and 582 of the 2008 Minnesota Statutes, particularly in 582.2, Subdivision 2. Mississippi: Lenders may foreclose on deeds of trusts or mortgages in default using either a judicial or a non-judicial foreclosure process. State law permits a deficiency judgment provided that the lender files for one within one year of the foreclosure sale date. If a mortgagee participates in foreclosure sale auction, his bid must pass a judicial standard of reasonableness. We classify Mississippi as a RECOURSE state. The relevant statutes are found in Section 89-1-305 of the Mississippi State Code. Recourse and Residential Mortgage Default: Evidence from US States 3181 Downloaded from rfs.oxfordjournals.org by guest on September 7, 2011 Missouri: Lenders may foreclose through either a judicial or a non-judicial procedure. The state has a statutory right of redemption, but the burden on the borrower is prohibitively heavy and this right can rarely be exercised. In the case of a non-judicial foreclosure sale, a separate court action must be filed to obtain a deficiency judgment but there are no other significant restrictions on obtaining a deficiency judgment. We classify Missouri as a RECOURSE state. The relevant statutes are found in the Missouri Revised Statutes, Chapter 141, Sections 400–590. Montana: Lenders may foreclose through either a judicial or a non-judicial procedure. Deficiency judgments are prohibited on purchase mortgages by Title 71, Chapter 1–232, of the Montana Code Annotated. Deficiency judgments are permitted on other types of residential mortgages only if the lender pursues judicial foreclosure. However, judicial foreclosure is often impractical because the grantor is entitled to a one-year right of redemption. The non-judicial foreclosure process is also substantially less complicated and costly. We classify Montana as a NON-RECOURSE state. The relevant statutes are found in Title 71, Chapter 1, of the Montana Code Annotated. Nebraska: Lenders may foreclose through either a judicial or a non-judicial procedure. Lenders may obtain a deficiency judgment. However, the borrower must receive credit for the fair market value of the property and the deficiency must be filed for within 90 days of the foreclosure sale by non-judicial foreclosure and within five years in cases of judicial foreclosure. We classify Nebraska as a RECOURSE state. The relevant statutes are found in the Nebraska Revised Statutes, Chapter 76-1013. Nevada: Lenders may foreclose through either a judicial or a non-judicial procedure. Usually, properties are foreclosed through a non-judicial procedure. A deficiency judgment can be obtained, but the borrower must receive credit for the greater of the fair market value of the property, as determined through a hearing, or the foreclosure sale price. The lender must file for a deficiency judgment within 90 days of the foreclosure sale. We classify Nevada as a RECOURSE state. The relevant statutes are found in the Nevada Revised Statutes, Chapters 40, 106, and 107. New Hampshire: Lenders may foreclose through either a judicial or a non-judicial procedure. Almost all properties are foreclosed non-judicially. There are no significant restrictions on deficiency judgments. We classify New Hampshire as a RECOURSE state. The relevant statutes are found in Title 38, Chapter 479, of the New Hampshire Revised Statutes. New Jersey: Lenders foreclose through a judicial process. State law permits deficiency judgments, but the borrower must be given credit for the fair market value of the property and the judgment must be brought within three months of the foreclosure sale. The pursuit of a deficiency judgment extends the redemption period from 10 days to 6 months. We classify New Jersey as a RECOURSE state. The relevant statutes are found in the New Jersey Permanent Statutes, Title 2A, Section 50. New Mexico: Lenders foreclose on residential properties through a judicial process. Deficiency judgments on mortgages and deeds of trust other than those used to finance low-income housing can be obtained, and there are no significant restrictions. We classify New Mexico as a RECOURSE state. The relevant statutes are found in Chapter 48, Articles 48-7-1 to 48-7-24 and Articles 48-10-1 to 48-10-21, of the New Mexico Statutes Annotated. New York: Lenders may foreclose through either a judicial or a non-judicial procedure, although non-judicial foreclosure is exceptionally rare. State law permits a deficiency judgment provided that the lender submits a request for a deficiency judgment within 90 days of filing the foreclosure suit. However, the borrower receives credit for the greater of the foreclosure sale price or the fair market value of the property. The judge usually sides with the borrower regarding the fair market value of the property. A typical deficiency judgment is relatively expensive. We classify New York as a RECOURSE state. The relevant statutes are found in Article 13 of the New York State Consolidated Laws. North Carolina: Lenders may foreclose through either a judicial or a non-judicial process. Chapter 45, Article 2B, Section 21.38, of the North Carolina General Statutes prohibits deficiency judgments on purchase mortgages. We classify purchase mortgages in North Carolina as NON-RECOURSE. Deficiency judgments are permitted on other types of residential mortgages, The Review of Financial Studies / v 24 n 9 2011 3182 Downloaded from rfs.oxfordjournals.org by guest on September 7, 2011 but the borrower has the right to contest the deficiency judgment in order to receive credit for the fair market value of the property. The deficiency judgment must be filed within one year. North Carolina law does not permit garnishment of wages to collect debt. We classify non-purchase mortgages in North Carolina as RECOURSE. The relevant statutes are found in Sections 21.36 and 21.38 of Article 2B in Chapter 45 of the North Carolina General Statutes. North Dakota: Lenders foreclose through a judicial process. Chapter 32-19-01 of the North Dakota Century Code prohibits deficiency judgments on residential properties. This provision applies to residential property with four or fewer units on up to 40 contiguous acres if at least one unit is owner-occupied. We classify North Dakota as a NON-RECOURSE state. Ohio: Lenders may foreclose only through judicial foreclosure. If the debt is greater than the foreclosure sales price plus reasonable expenses, a deficiency judgment is automatic. However, lenders have only two years to collect on the deficiency. We classify Ohio as a RECOURSE state. The relevant statutes are found in the Ohio Revised Code, Section 2329.08. Oklahoma: Lenders may foreclose through either judicial or non-judicial foreclosure. The optimum timeframe for non-judicial foreclosure is 201 days. Lenders may only receive a deficiency judgment if they pursue non-judicial foreclosure, and the borrower must receive credit for the greater of the fair market value or the foreclosure sale price. The lender must file for a deficiency judgment within 90 days of the foreclosure sale. We classify Oklahoma as a RECOURSE state. The relevant statute is Title 12, Chapter 12, Section 686, of the Oklahoma Statutes Citationized. Oregon: Lenders may foreclose through either a judicial or a non-judicial procedure. Lenders can generally not obtain a deficiency judgment on a residential property. We classify Oregon as a NON-RECOURSE state. Pennsylvania: Lenders foreclose through a judicial procedure. Pennsylvania law permits the lender to file for a deficiency judgment in a suit separate from the foreclosure, but the borrower must receive credit for the fair market value of the property. The deficiency suit must be brought within six months of the foreclosure sale. We classify Pennsylvania as a RECOURSE state. The relevant statute is the Pennsylvania Deficiency Judgment Act, Chapter 81, Section 8103, of the Pennsylvania Consolidated Statutes. Rhode Island: Lenders may foreclose through either a judicial or a non-judicial procedure. Deficiency judgments can be obtained, and there are no significant restrictions. We classify Rhode Island as a RECOURSE state. The relevant statutes are found in Chapter 34-27 of the Rhode Island General Laws. South Carolina: Lenders foreclose through a judicial procedure. State law permits deficiency judgments subject to the restriction that the borrower must receive credit for the fair market value of the property. In such a circumstance, the borrower, judge, and lender all hire appraisers to determine the fair market value of the property. We classify South Carolina as a RECOURSE state. The relevant statutes are found in Title 29, Chapter 3, Article 7, of the South Carolina Code of Laws. South Dakota: Lenders may foreclose through either a judicial or a non-judicial procedure. State law permits deficiency judgments provided the borrower is credited for the fair market value of the property. We classify South Dakota as a RECOURSE state. The relevant statutes are found in Chapter 21-47 of the South Dakota Codified Laws. Tennessee: Lenders may foreclose through either a judicial or a non-judicial procedure, although lenders seldom use the judicial foreclosure process. State law permits deficiency judgments without significant restrictions. We classify Tennessee as a RECOURSE state. The relevant statutes for non-judicial foreclosure are Title 21, Chapter 1, Section 803, of the Tennessee Code. Texas: Lenders may foreclose through either a judicial or a non-judicial procedure, but a lender must foreclose on a home equity loan through a judicial foreclosure process. State law permits deficiency judgments subject to the borrower receiving credit for the fair market value of the property. However, Texas has a nearly unlimited homestead exemption, such that lenders have less recourse on mortgages backed by investment properties if the borrower’s primary residence is Recourse and Residential Mortgage Default: Evidence from US States Appendix B: Data Description B.1 Sample restrictions We restrict our analysis to mortgages with constant principal and interest, ARMs, or graduated payment mortgages (GPM) (the variable INT TYPE takes a value of 1, 2, or 5, respectively). Also, we restrict the analysis to mortgages taken out for purchase or refinance (the PURPOSE TYPE MCDASH variable takes the value of 1 for purchase, 2 for refinance (cash out), 3 for refinance (no cash out), or 5 for refinance (unknown cash)). We drop mortgages taken out for home improvement, debt consolidation, education, medical, or “other” reasons. The analysis is limited to first mortgages (the variable MORT TYPE takes the value of 1 for first mortgage, or 4 for a first mortgage grade of “B” or “C”). We restrict the sample to single-family residences, townhouses, or condos (PROP TYPE=1 or C). 3183 Downloaded from rfs.oxfordjournals.org by guest on September 7, 2011 also in Texas. We classify Texas as a RECOURSE state. The relevant statutes are found in Title 5, Section 51, of the Texas Statutes. Utah: Lenders may foreclose through either a judicial or a non-judicial procedure. State law permits deficiency judgments without significant restrictions. We classify Utah as a RECOURSE state. The relevant statutes are found in Title 38, Chapters 1–16, and Title 57, Chapter 1, of the Utah Code. Vermont: Lenders may foreclose through either a judicial or, if the mortgage contains a power of sale clause, a non-judicial procedure. The norm is judicial foreclosure. State law permits deficiency judgments with no significant restrictions. We classify Vermont as a RECOURSE state. The relevant statutes are found in Title 12, Chapter 163, of the Vermont Statutes. Virginia: Lenders may foreclose through either a judicial or a non-judicial process. State law permits deficiency judgments with no significant restrictions. We classify Virginia as a RECOURSE state. The relevant statutes are found in Title 8.9A, Part 6, and Title 55, Chapter 4, of the Code of Virginia. Washington: Lenders may foreclose through either a judicial or a non-judicial process. If the lender wishes to pursue a deficiency judgment, however, it must pursue judicial foreclosure, and pursuit of a deficiency judgment triggers a 12-month right of redemption. Furthermore, the judicial foreclosure process is substantially more time consuming than the non-judicial process. In addition, deficiency judgments cannot be obtained if the property has been abandoned for six months or more, which we view as one way a strategic defaulter could relatively easily evade a deficiency judgment. We classify Washington as a NON-RECOURSE state. The relevant statutes are found in Title 61, Chapter 61–12, of the Revised Code of Washington. West Virginia: Lenders may foreclose through either a judicial or a non-judicial process. West Virginia permits deficiency judgments without significant restrictions. We classify West Virginia as a RECOURSE state. The relevant statutes are found in Articles 1 and 16 of Chapter 38 of the West Virginia Code. Wisconsin: Lenders foreclose through a non-judicial process. A deficiency judgment must be filed at the time the foreclosure action starts. A waiver of a deficiency judgment may reduce a redemption period from 12 months to 6 months, and a redemption period of 6 months to 3 months. The redemption period depends on a number of characteristics, including parcel size. We classify Wisconsin as a NON-RECOURSE state. The relevant statutes can be found in Wisconsin Statutes and Annotations, Chapter 846. Wyoming: Lenders may foreclose through either a judicial or a non-judicial process. At a foreclosure sale, the lender generally bids the lesser of the debt owed or the fair market value of the property. State law permits deficiency judgments without significant restrictions. We classify Wyoming as a RECOURSE state. The relevant statutes are found in Title 34, Chapter 4, of the Wyoming Statutes. The Review of Financial Studies / v 24 n 9 2011 B.2 Variable definitions B.2.1 Definition of default. We consider the loan as defaulted if it is terminated in one of the following ways: by REO sale, by short sale, by payoff out of foreclosure, by payoff out of bankruptcy or serious delinquency, or by liquidation to termination. We do not count terminations by voluntary payoff or by a loan transfer from a servicer as defaults. The default month is the first month in which the defaulted loan is reported as being in foreclosure, in REO proceedings, or under liquidation, whichever comes first (MBA STAT variable takes a value of F, R, or L, respectively). In addition, if the loan is terminated by default without a loan status reported as any of the three mentioned above, the default month is the month when the loan is reported as paid off. Finally, if TERMINATION TYPE=8, we count the loan as defaulted because the FORECLOSURE TYPE for these variables is non-zero, indicating that there was a foreclosure. Less than 0.1% of loans are terminated with TERMINATION TYPE=8. B.2.2 Default type. If a loan goes from being in foreclosure to being an REO loan, we treat it as a foreclosure. That is, we define a foreclosure as any loan for which MBA STAT=F prior to it being any other MBA STAT. B.2.3 Default option. For the current principal balance amount, we use variable PRIN BAL B.2.4 Prepay option variables. The ongoing contract rate on the mortgage is contained in the variable CUR INT RATE. The market mortgage rate is a contract rate on the composite of all conventional mortgage loans (fixed and adjustable rate) from the Finance Board’s Monthly Survey of Rates and Terms on Conventional Single-family Non-farm Mortgage Loans. The survey collects information on fully amortized conventional mortgage loans used to purchase single-family, non-farm homes. Mortgage loans insured by the Federal Housing Administration or guaranteed by the Veterans Administration are excluded. Loans used to refinance houses and non-amortized and balloon loans are also excluded. The data are available in Table 17 at http://www.fhfb.gov/ Default.aspx?Page=8&Top=4. B.2.5 Trigger events. State divorce rates are available on an annual basis for most years in our sample from the Division of Vital Statistics at the CDC’s National Center for Health Statistics. The data are available at http://www.cdc.gov/nchs/data/nvss/Divorce%20Rates%2090%2095%20 and%2099-07.pdf. We interpolate the values for 1997 and 1998 from the 1995 and 1999 values and use the 2007 value for 2008. B.2.6 Loan level characteristics • Ln loan age in months from the closing date to the contemporaneous month • LTV at origination • An indicator variable if the loan is interest only at origination (IO FLAG) 3184 Downloaded from rfs.oxfordjournals.org by guest on September 7, 2011 AMT (the balance the borrower owes on the loan). For the cost of purchase, we use variable ORIG AMT (original loan amount). Loans for which the principal balance amount at the time of default (which is described below) is zero or missing and cannot be imputed from up to two previous months are dropped from the analysis. To calculate ki , we use the loan closing date (CLOSE DT; as is used by McDash). The OFHEO provides a quarterly (not seasonally adjusted) measure of the House Price Index by state (http://www.ofheo.gov/hpi download.aspx). We use the all-transactions index. As the OFHEO provides A and B in quarters, we convert months since origination into quarters since origination. We also construct monthly values of HPI i,t by linearly interpolating from the quarterly values, attributing the quarterly value to the second month of the quarter. Recourse and Residential Mortgage Default: Evidence from US States • An indicator variable if the loan is an option ARM (INT TYPE) • An indicator variable if the loan is a jumbo (JUMBO FLG) • An indicator variable if the loan is a refinance mortgage (PURPOSE TYPE MCDASH) • The borrower’s FICO score at origination (FICO ORIG - original FICO score, available from 8/1997) References Adelino, M., K. Gerardi, and P. S. Willen. 2009. Why Don’t Lenders Renegotiate More Home Mortgages? Redefaults, Self-cures, and Securitization. Working Paper 09-4, Federal Reserve Bank of Boston. Ambrose, B. W., R. J. Buttimer Jr., and C. A. Capone. 1997. Pricing Mortgage Default and Foreclosure Delay. Journal of Money, Credit, and Banking 29:314–25. Ambrose, B. W., C. A. Capone, and Y. Deng. 2001. Optimal Put Exercise: An Empirical Examination of Conditions for Mortgage Foreclosure. Journal of Real Estate Finance and Economics 23:213–34. Brueggeman, W. B., and J. D. Fisher. 2011. Real Estate Finance and Investments. 14th Ed. New York: McGraw-Hill Irwin. Calhoun, C. A. 1996. OFHEO House Price Indexes: HPI Technical Description. Working Paper, OFHEO. Campbell, J. Y., S. Giglio, and P. Pathak. Forthcoming. Forced Sales and House Prices. American Economic Review. Case, K. E., and R. J. Shiller. 1987. Prices of Single-family Homes Since 1970: New Indexes for Four Cities. New England Economic Review Sept./Oct.:45–56. Clauretie, T. M. 1987. The Impact of Interstate Foreclosure Cost Differences and the Value of Mortgages on Default Rates. American Real Estate and Urban Economics Association Journal 15:152–67. Corbae, D., and E. Quintin. 2010. Mortgage Innovation and the Foreclosure Boom.Working Paper, University of Texas at Austin. Crawford, G. W., and E. Rosenblatt. 1995. Efficient Mortgage Default Option Exercise: Evidence from Loss Severity. Journal of Real Estate Research 10:543–55. Deng, Y., J. M. Quigley, and R. Van Order. 2000. Mortgage Terminations, Heterogeneity, and the Exercise of Mortgage Options. Econometrica 68:275–307. Experian-Oliver Wyman. 2009. Market Intelligence Report on Strategic Defaults. Mimeo. Feldstein, M. 2008. How to Help People Whose Home Values Are Underwater—The Economic Spiral Will Get Worse Unless We Do Something about Negative Equity. Wall Street Journal, November 18. Foote, C. L., K. S. Gerardi, and P. S. Willen. 2008. Negative Equity and Foreclosure: Theory and Evidence. Journal of Urban Economics 64:234–45. Foote, C. L., K. S. Gerardi, L. Goette, and P. S. Willen. 2009. Reducing Foreclosures: No Easy Answers. NBER Macroeconomics Annual 24:89–138. Guiso, L., P. Sapienza, and L. Zingales. 2010. The Determinants of Attitudes Toward Strategic Default on Mortgages. Working Paper, University of Chicago. Jones, L. D. 1993. Deficiency Judgments and the Exercise of the Default Option in Home Mortgage Loans. Journal of Law and Economics 36:115–38. Kau, J. B., D. C. Keenan, and T. Kim. 1994. Default Probabilities for Mortgages. Journal of Urban Economics 35:278–96. 3185 Downloaded from rfs.oxfordjournals.org by guest on September 7, 2011 Bhutta, N., J. Dokko, and H. Shan. 2010. The Depth of Negative Equity and Mortgage Default Decisions. Finance and Economics Discussion Series Working Paper 2010–35, Federal Reserve Board of Governors. The Review of Financial Studies / v 24 n 9 2011 Larsen, J. E., B. Carey, and C. Howell Carey. 2007. Real Estate. Hoboken, NJ: Wiley. Leland, J. 2008. Facing Default, Some Walk Out on New Homes. New York Times, February 29. Ling, D. C., and W. R. Archer. 2008. Real Estate Principles: A Value Approach. 2nd Ed. New York: McGrawHill Irwin. Lunn, M., and D. McNeil. 2005. Applying Cox Regression to Competing Risks. Biometrics 51:524–32. Pence, K. M. 2006. Foreclosing on Opportunity: State Laws and Mortgage Credit. Review of Economics and Statistics 88:177–82. USFN. 2004. The National Mortgage Servicer’s Reference Directory. Tustin, CA: USFN. White, M. J. 1998. Why It Pays to File for Bankruptcy: A Critical Look at the Incentives under U.S. Personal Bankruptcy Law and a Proposal for Change. University of Chicago Law Review 65:685–732. Downloaded from rfs.oxfordjournals.org by guest on September 7, 2011 3186